UNIVERSITY OF OULU P .O. Box 8000 F I -90014 UNIVERSITY OF OULU FINLAND

A C T A U N I V E R S I T A T I S O U L U E N S I S

University Lecturer Tuomo Glumoff

University Lecturer Santeri Palviainen

Postdoctoral research fellow Sanna Taskila

Professor Olli Vuolteenaho

University Lecturer Veli-Matti Ulvinen

Planning Director Pertti Tikkanen

Professor Jari Juga

University Lecturer Anu Soikkeli

Professor Olli Vuolteenaho

Publications Editor Kirsti Nurkkala

ISBN 978-952-62-1956-1 (Paperback)ISBN 978-952-62-1957-8 (PDF)ISSN 0355-3213 (Print)ISSN 1796-2226 (Online)

U N I V E R S I TAT I S O U L U E N S I SACTAC

TECHNICA

U N I V E R S I TAT I S O U L U E N S I SACTAC

TECHNICA

OULU 2018

C 665

Ayotunde Oluwaseun Laiyemo

HIGH SPEED MOVING NETWORKS IN FUTURE WIRELESS SYSTEMS

UNIVERSITY OF OULU GRADUATE SCHOOL;UNIVERSITY OF OULU,FACULTY OF INFORMATION TECHNOLOGY AND ELECTRICAL ENGINEERING;CENTRE FOR WIRELESS COMMUNICATIONS

C 665

AC

TAA

yotunde Oluw

aseun Laiyemo

C665etukansi.fm Page 1 Thursday, May 17, 2018 8:45 AM

ACTA UNIVERS ITAT I S OULUENS I SC Te c h n i c a 6 6 5

AYOTUNDE OLUWASEUN LAIYEMO

HIGH SPEED MOVING NETWORKS IN FUTURE WIRELESS SYSTEMS

Academic dissertation to be presented with the assent ofthe Doctoral Training Committee of Technology andNatural Sciences of the University of Oulu for publicdefence in the OP auditorium (L10), Linnanmaa, on 15August 2018, at 12 noon

UNIVERSITY OF OULU, OULU 2018

Copyright © 2018Acta Univ. Oul. C 665, 2018

Supervised byProfessor Matti Latva-ahoDocent Pekka Pirinen

Reviewed byProfessor Markus RuppProfessor Tommy Svensson

ISBN 978-952-62-1956-1 (Paperback)ISBN 978-952-62-1957-8 (PDF)

ISSN 0355-3213 (Printed)ISSN 1796-2226 (Online)

Cover DesignRaimo Ahonen

JUVENES PRINTTAMPERE 2018

OpponentProfessor Thomas Kurner

Laiyemo, Ayotunde Oluwaseun, High speed moving networks in future wirelesssystems. University of Oulu Graduate School; University of Oulu, Faculty of Information Technologyand Electrical Engineering; Centre for Wireless CommunicationsActa Univ. Oul. C 665, 2018University of Oulu, P.O. Box 8000, FI-90014 University of Oulu, Finland

Abstract

This thesis concentrates on evaluating and improving the throughput performances of mobileusers in high speed vehicles. In particular, high speed train (HST) scenarios are considered.Emphasis is placed on practical designs and methods that take into account distinctive HSTcharacteristics. A two-hop communication link, i.e., base station (BS)-to-HST and HST-to-onboard users (OBUs) is adopted. The main target is to improve the throughput performance onthe BS-to-HST communication link, which is assumed to be the main bottleneck in the wholecommunication link, since the HST-to-OBU communication link is assumed to have good channelquality due to the short link distance with relatively stationary OBUs. The algorithms developedare assessed through link and system level simulations.

A theoretical and practical study of the throughput maximization problem in a single and multi-cell multiple-input multiple-output orthogonal frequency-division multiplexing (MIMO-OFDM)train scenario are considered with and without cooperation between train carriages. Two low-complexity transmission schemes based on simple antenna selection (AS) methods with spatialmultiplexing (SM) are proposed. The simulation results demonstrate that large antenna arrays withAS and SM transmission strategies have the potential to significantly improve the throughput ofthe BS-to-train link in HST scenarios.

Resource sharing methodologies between the moving relay nodes (MRNs) on the HST andground macro users (GMUs) were also studied in a multi-cell MIMO-OFDM train scenario. Directapplication of existing resource scheduling methods will not be appropriate to efficiently andfairly share resources, since the MRNs and the GMUs have different processing capabilities.Hence, two hybrid resource scheduling methods are analyzed in conjunction with joint and disjointresource management. The tradeoff between the number of MRNs and receive antennas thatshould be installed on an HST was also examined in the context of throughput performance andcapital expenditure. Results show that joint scheduling does not provide the best overallperformance and there is a need to schedule each group of mobile terminals (MTs) separately.

Finally, the feasibility of the use of higher frequency bands (HFBs) was examined in HSTscenarios. A timer-based beam selection scheme for HST, which does not require any training timeto select the appropriate beam is also proposed. The proposed beam selection scheme (PBSS)displays a close performance to the ideal singular value decomposition (SVD) scheme.

Keywords: antenna selection, high mobility communication, higher frequency bandbeamforming, LTE-A, MIMO-OFDM, resource scheduling, spatial multiplexing,system level simulation

Laiyemo, Ayotunde Oluwaseun, Langattomat verkot nopeasti liikkuvilleterminaaleille tulevaisuuden tietoliikennejärjestelmissä. Oulun yliopiston tutkijakoulu; Oulun yliopisto, Tieto- ja sähkötekniikan tiedekunta; Centre forWireless CommunicationsActa Univ. Oul. C 665, 2018Oulun yliopisto, PL 8000, 90014 Oulun yliopisto

Tiivistelmä

Tämä väitöskirja keskittyy mobiilikäyttäjien tiedonsiirtonopeuksien arviointiin ja parantamiseennopeasti liikkuvissa kulkuneuvoissa. Työ käsittelee erityisesti tiedonsiirtoa suurnopeusjunissa.Työssä korostetaan käytännön menetelmiä, jotka ottavat huomioon nopeasti liikkuvien junientiedonsiirron erityispiirteet. Työssä käytetään kahden hypyn linkkimallia, jossa tiedonsiirto kul-kee tukiasemalta junaan ja junasta käyttäjälle, joka on junassa. Päätavoite on parantaa datanope-uksia tukiaseman ja junan välisessä tiedonsiirtolinkissä, jonka uskotaan olevan suurin pullon-kaula koko tiedonsiirtolinkissä, koska junan ja lähes paikallaan olevan käyttäjän välinen kanavavoidaan olettaa hyvälaatuiseksi linkin lyhyyden vuoksi. Kehitettyjen algoritmien suorituskykyäarvioidaan linkki- ja järjestelmätason simulaatioilla.

Työssä tutkitaan tiedonsiirtonopeuden maksimointiongelmaa teoreettisella ja käytännöntasolla yhden ja usean solun MIMO OFDM junaskenaarioissa, joissa junan vaunut tekevät taieivät tee yhteistyötä. Työssä esitetään kaksi alhaisen kompleksisuuden lähetysmenetelmää, jotkahyödyntävät yksinkertaista antennin valintamenetelmää ja tilatason multipleksointia. Simulointi-tulokset osoittavat, että suuret antenniryhmät, jotka hyödyntävät näitä lähetysmenetelmiä, voi-vat parantaa merkittävästi tiedonsiirtonopeutta tukiasemalta junaan päin.

Työssä tutkitaan myös resurssien jakomenetelmiä liikkuvien junassa olevien releiden ja maa-tason makrokäyttäjien välillä monen solun MIMO-OFDM junaskenaariossa. Nykyisten resurs-sinhallintamenetelmien käyttö ei ole suoraan mahdollista tehokasta ja oikeudenmukaista resurs-sien jakoa, koska releillä ja makrokäyttäjillä on erilaiset prosessointikyvyt. Tämän vuoksi työs-sä analysoidaan kahta hybridimenetelmään resurssien skeduloinnille. Tutkimukset selventävättasapainoa releiden lukumäärän ja junaan asennettavien vastaanotinantennien välillä tiedonsiir-tonopeuden ja kustannusten osalta. Tulokset osoittavat, että yhteinen resurssien jako ei saavutaparasta suorituskykyä, eikä ole tarvetta ajoittaa jokaista matkaviestinterminaaliryhmää erikseen.

Lopuksi työssä tutkitaan korkeampien taajuusalueiden soveltuvuutta tiedonsiirtoon suurno-peusjunissa. Työssä ehdotetaan ajastinpohjaista keilanvalintamenetelmää, joka ei vaadi opetus-jaksoa sopivan keilan valintaan. Ehdotetun menetelmän saavuttama suorituskyky on lähellä ide-aalisen singulaariarvohajotelmaa hyödyntävän menetelmän suorituskykyä.

Asiasanat: antennin valinta, järjestelmätason simulointi, korkeampien taajuuksienkeilanmuokkaus, korkean mobiliteetin tiedonsiirto, LTE-A, MIMO-OFDM, resurssienskedulointi, tilatason multipleksointi

Dedicated to my family

8

Preface

This doctoral thesis consists of research activities carried out at the Centre for WirelessCommunications (CWC), at the University of Oulu in Finland. The research describedherein was conducted under the supervision of Professor Matti Latva-aho and AdjunctProfessor Pekka Pirinen. From the start of my doctoral studies, I have encounteredmany individuals and organizations who have contributed in one way or the other to thesuccess of my doctoral program. For this, I would like to express my sincere gratitude,which extends well beyond the names listed here, which has been limited for sake ofbrevity.

First, I would like to express my appreciation for my supervisor, Professor MattiLatva-aho, for giving me the opportunity to work in his highly esteemed research unit.I am also grateful for his endless support, enthusiasm and concerns, not only in myresearch work, but also for my welfare. I wish to express my deep gratitude to myco-supervisor, Adjunct Professor Pekka Pirinen, for his meticulous guidance, knowledgeand encouragement during my studies. My gratitude also goes to my advisor, DoctorHarri Pennanen, for his diligent guidance, ingenuity and tremendous support throughoutmy studies.

I would like to appreciate the efforts of my follow-up group members, ProfessorMarkus Katz, Assistant Professor Le-Nam Tran and Doctor Zaheer Khan for theirinsightful advice and words of encouragement. I also wish to thank the reviewers of thethesis, Professor Markus Rupp from Technische Universitat Wien Austria, and ProfessorTommy Svensson from Chalmers University of Technology, for accepting to review mythesis and providing constructive comments.

I would like to thank Adjunct Professor Hirley Alves, particularly for constantlykeeping me informed on the latest articles in my area of research, together with my pastand present office colleagues (TS414) for their help and friendship during my doctoralstudies. They are Doctor Francesco Pantisano, Associate Professor Pedro Nardelli,Doctor Keeth Jayasinghe, Doctor Dan Nguyen, Doctor Petri Luoto, Doctor SumuduSamarakoon, Doctor Manosha Kapuruhamy, Mohammed Elbamby, Kien Vu, Chen-FengLiu, and Mohammed Khairy. I am also grateful to the administrative staff at CWC forcreating a friendly and inspiring work environment.

9

During my doctoral studies at CWC, I have been privileged to work on the fol-lowing projects and I wish to acknowledge their contribution to the success of mydoctoral studies; Local Connectivity and Cross-Layer Design for Future BroadbandMobile Systems (LOCON), Solutions for Capacity Crunch in Wireless Access withFlexible Architectures (CRUCIAL), 5G radio access solutions to 10 GHz and beyondfrequency bands (5Gto10G) and the Satellite and Terrestial network for 5G (SAT5G)project. The projects were funded by the European Commission, the Finnish FundingAgency for Technology and Innovation (Tekes), Nokia, Huawei Technologies, AniteTelecoms/Keysight Technologies, Broadcom Communications Finland, and ElektrobitWireless Communications. I also wish to acknowledge the University of Oulu Scholar-ship foundation for the scholarship grant offered to me. This thesis has been financiallysupported by Academy of Finland 6Genesis Flagship (grant 318927).

I wish to express my thanks and appreciation to my dad Joshua Laiyemo and mymum Comfort Laiyemo for their unconditional love and spiritual support throughoutmy life. I would also like to thank my brother Ayodele Laiyemo and my sister FolukeLaiyemo for their love, constant support and encouragement. To my wife, FunkeLaiyemo, I express my deepest gratitude for her love, support, and faith in me. And tomy daughter, Darasimi Laiyemo, thank you for aspiring to be a good girl.

Above all, I give all the glory to the King of kings, the author of my life for his graceand favor. He has made things fall into place for me in so many wonderful ways duringthe course of my life, particularly during my stay in Finland.

Oulu, 15.08.2018 Ayotunde Oluwaseun Laiyemo

10

Abbreviations

3GPP 3rd Generation Partnership Project4G Fourth Generation5G Fifth Generation5G-R 5G Communication System for RailwayACLR Adjacent Channel Leakage RatioAoA Angle of ArrivalAoD Angle of DepartureAS Antenna SelectionBEM Basis Expansion ModelBER Bit Error RateBS Base StationCDD Cyclic Delay DiversityCDF Cumulative Distribution FunctionCDT Count Down TimerCQI Channel Quality IndicatorcrX2 Cooperative relay X2CSI Channel State InformationD2D Device to DeviceDCP Difference of Convex Function ProgramDFT Discrete Fourier TransformsESPRIT Estimation of Signal Parameters via Rotational Invariance TechniqueFBMC Filter Bank Multiple AccessFDD Frequency Division DuplexingFEP Frame Error ProbabilityGMU Ground Macro UserHARQ Hybrid Automatic Repeat RequestHD High DefinitionHFB Higher Frequency BandHH HouseholderHST High Speed TrainICeI Inter cell interference

11

ICI Inter-Carrier-InterferenceIMT-A International Mobile Telecommunications AdvancedISI Inter-Symbol-InterferenceITU-R International Telecommunication Union RadiocommunicationsL2S Link to SystemLMMSE Linear Minimum Mean Square ErrorLOS Line-Of-SightLTE Long-Term EvolutionLTE-A Long-Term Evolution-AdvancedMCS Modulation and Coding SchemeMGMA Multi-Group Multi-AntennaMIESM Mutual Information Effective SINR MetricMIMO Multiple-Input Multiple-OutputMMSE Minimum Mean Square ErrormmWave Millimeter WaveMPF Modified Proportional FairMRN Moving Relay NodeMS Mobile StationMSE Mean Square ErrorMT Mobile TerminalMUSIC Multiple Signal ClassificationNF Noise FigureNLOS Non Line-of-SightOBU Onboard UserOFDM Orthogonal Frequency Division MultiplexingPBSS Proposed Beamforming Selection SchemePF Proportional FairPMI Precoding Matrix IndicatorQFI QoS Fulfilment InformationQoS Quality of ServiceRAU Remote Antenna UnitRF Radio FrequencyRI Rank IndicatorRRC Radio Resource ControlRRMR Round Robin with Max Rate

12

RRU Radio Remote UnitRSS Received Signal StrengthSCA Successive Convex ApproximationSIC Successive Interference CancellationSINR Signal to Interference plus Noise RatioSM Spatial MultiplexingSNR Signal to Noise RatioSSCM Statistical Spatial Channel ModelSVD Singular Value DecompositionTTI Transmission Time IntervalTTT Time to TriggerV2I Vehicular to InfrastructureV2V Vehicle to VehicleVPL Vehicular Penetration LossWiFi Wireless FidelityM total number of MRNswc,l receive filter at the cth subcarrier on the lth streamyc,m receive signal vector at the mth MRN for the cth subcarriernc,m additive complex white Gaussian noise vectortc,m,l auxiliary variable at the mth MRN for the cth subcarrier on the lth streamCm,r(t) capacity of the rth scheduling block for the mth MRN on the t th TTIDc symbol for cyclic delay diversity (CDD)Hc channel matrix at the cth subcarrierHc,i channel matrix at the cth subcarrier for the ith RF chainHc,m channel matrix from BS to the mth MRN on the cth subcarrierHc,k channel matrix from interfering BS k to the MRN systemsc,m,l data symbol for the lth stream of the mth MRNU symbol for discrete Fourier transforms (DFT)Hc estimate of the channel matrix Hc

Rc estimate of the interference plus noise covariance matrix Rc

TCP length of cyclic prefixTu length of the useful symbolWc LMMSE filter vectorfD maximum Doppler shiftNtmax maximum number of transmit antennas

13

Γmin minimum SNR/SINR for successful transmissionεc,m,l MSE at the mth MRN for the cth subcarrier on the lth streamFc normalized precoding matrixNcar number of carriages on the trainLm number of data streams given to the mth MRNQ number of GMUsNr number of MRN receive antennasMb number of MRNs jointly associated to BS b

Nsym number of OFDM symbolsNrg number of GMU receive antennasNRB number of resource blocksNr f number of RF chainsnmax

SB number of scheduling blocks allocated to an MT at a timeNSB number of scheduling blockC number of subcarriersCr number of subcarriers in the rth scheduling blockNt number of transmit antennas#MRN∗ optimum number of MRNsN∗r optimum number of receive antennasm∗ optimum selected MTr∗ optimum selected scheduling blockPL( fc,d) pathloss for a given center frequency and separation distancePc power vector with powers allocated to each L streamsMc,k precoder matrix used at BS k to transmit Lk (interfering) data streamswc,m,l receive filter for the lth stream of the mth MRNyc received signal vector for the cth subcarrieryc,i received signal vector at the cth subcarrier for the ith RF chainΓc received SNR/SINR at the cth subcarrierΓc,l received SINR at the cth subcarrier on the lth streamΓc,m,l received SINR of the lth stream at the mth MRNB set of BSs within the networkU set of GMU/MRNs associated with the same BSP set of possible transmission precoding matrixR set of possible transmission rankW set of receive beamforming vectors

14

M set of transmit beamforming vectorsβ SNR efficiency factorRb spatial correlation matrix at the BSRr spatial correlation matrix at the MRN system∆ f subcarrier spacingα system bandwidth efficiency factorRtar target rateTNt throughput with selected number of active transmit antennasTr, f throughput with selected precoder matrix f and rank r

Tslot time slotRtot(b) total achievable rate in the bth cellRgmu(b) total GMU achievable rate in the bth cellRmrn(b) total MRN achievable rate in the bth cellL total number of data streams transmitted to all M MRNsK total number of the interfering BSsT total number of TTIsP total transmit powerGb

t transmit antenna gain from the bth serving BSGb′

t transmit antenna gain from the b′th interfering BSPb

t transmit power from the bth serving BSPb′

t transmit power from the b′th interfering BSmn,i transmit beamforming vector of the nth beam for the ith RF chainmc,m,l transmit precoding vector for the lth stream of the mth MRNmc,l transmit precoding vector of the lth stream at the cth subcarriersc transmit signal vectorsn,i transmit signal vector at the nth beam on the ith RF chainB transmission bandwidthpc,l transmission power for the lth streamMc unnormalized transmit precoding matrix for the desired data streamsdiag(x) diagonal matrix with elements of vector x on the main diagonalE(.) statistical expectation‖x‖2 Euclidean norm of vector x

|x| absolute value of scalar x

C set of complex numbersR set of real numbers

15

16

Contents

AbstractTiivistelmäPreface 9Abbreviations 11Contents 171 Introduction 19

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

1.2 Challenges in high mobility scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1.2.1 Fast time varying fading channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

1.2.2 Large and frequent resource handovers . . . . . . . . . . . . . . . . . . . . . . . . . . .21

1.2.3 Presence of vehicular penetration loss . . . . . . . . . . . . . . . . . . . . . . . . . . . 22

1.3 Network design approach and assumptions for HST communication . . . . . . 22

1.3.1 Network architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

1.3.2 Mobility management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .25

1.3.3 Channel estimation and feedback delay . . . . . . . . . . . . . . . . . . . . . . . . . . 26

1.4 Literature review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.4.1 Transmission schemes for HSTs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

1.4.2 Resource scheduling approach for HSTs . . . . . . . . . . . . . . . . . . . . . . . . . 29

1.4.3 Higher frequency bands for HSTs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

1.5 Objectives, contribution, and outline of the thesis . . . . . . . . . . . . . . . . . . . . . . . 32

1.6 The author’s contribution to the publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

2 Transmission strategies for throughput maximization in HSTcommunications 352.1 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.1.1 Theoretical single-cell scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2.1.2 Practical multi-cell scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

2.2 Transmission techniques from theoretical perspective . . . . . . . . . . . . . . . . . . . . 40

2.2.1 MRN non-cooperative mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

2.2.2 MRN cooperative mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

2.3 Transmission techniques from practical perspective. . . . . . . . . . . . . . . . . . . . . .43

2.3.1 LTE precoding schemes applied to high speed train scenario . . . . . . . 44

17

2.3.2 Practical transmission schemes based on antenna selection andspatial multiplexing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

2.4 Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512.4.1 Simulator description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 512.4.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

2.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643 Resource management in HST communications 65

3.1 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 653.2 Resource scheduling approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66

3.2.1 Resource scheduling problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.2.2 Resource allocation methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

3.3 Optimum cooperative MRN deployment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .713.4 Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74

3.4.1 Simulator description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 743.4.2 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

3.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824 Beamforming at higher frequency bands for HST communications 83

4.1 Suitability of higher operating frequencies for railway communications. . . .834.1.1 The railway deployment scenario . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 834.1.2 HFB frame structure for HST and effect of large bandwidth . . . . . . . . 85

4.2 System model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 894.3 Beamforming for higher operating frequencies on railway networks . . . . . . . 89

4.3.1 Problem formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 904.3.2 Relationship between array gain and beamwidth . . . . . . . . . . . . . . . . . . 914.3.3 Impact of BS-HST link distance on the angle of arrival . . . . . . . . . . . . 924.3.4 Impact of velocity estimation error on the angle of arrival . . . . . . . . . .944.3.5 High speed train beam selection scheme . . . . . . . . . . . . . . . . . . . . . . . . . 95

4.4 Performance evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 964.4.1 HFB feasibility evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 984.4.2 Proposed beamforming scheme evaluation . . . . . . . . . . . . . . . . . . . . . . 103

4.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1075 Conclusion 109References 113Appendices 121

18

1 Introduction

This introductory chapter provides an overview of the thesis by first giving the relatedbackground, assumptions and a review of existing literature. In Section 1.1, themotivation for the need to improve the quality of service (QoS) for high mobility mobileterminals (MTs) is presented. The challenges encountered in high mobility scenarios areintroduced in Section 1.2. Section 1.3 describes the assumptions made and networkdesign approach chosen for the high speed train (HST) communication network. Section1.4 reviews previous related studies based on transmission schemes, resource allocation,and the use of higher frequency bands (HFBs) for HST communication networks.The objectives and outline of the thesis are described in Section 1.5 and the author’scontributions to the publications are given in Section 1.6.

1.1 Motivation

With the increased capabilities of portable mobile devices, global mobile data traffic andrates have been undergoing an exponential increase. For instance, mobile data trafficgrew by 63% and connection speeds increased by 70.6% in 2016 alone, with smartphoneusage accounting for 89% of the mobile data traffic [1]. In essence, a typical smartphonegenerates about 48 times more mobile data traffic than the typical basic cell phone,which is becoming extinct. Smartphones are handy, and studies have shown that about41% of passengers on public transportation systems use their smart phones onboard[2]. Hence, a large proportion of the mobile data traffic is generated at high mobilityas public transportation in most metropolitan areas is ever growing. To keep pacewith the huge growth in the generated mobile traffic and the large proportion of usersaccessing the internet at high mobility, the next generation mobile networks, i.e., the fifthgeneration (5G) must surpass the fourth generation’s (4G) long-term evolution-advanced(LTE/LTE-A) w.r.t. the data rate and capacity [3] by a great deal. Surpassing the 4Grequirements implies that 5G should be highly flexible and configurable to addressspecific use cases [4] as a result of the inherent diverse 5G requirements, of theserequirements, "broadband access in public transport" [5] is one of the use cases underconsideration. Therefore, it is important to study and develop schemes to meet the highrequirements for 5G networks on a case by case basis. One key concept is for 5G mobilesystems to broaden the spectrum to higher frequency bands (between 6 and 300 GHz) to

19

support high data rates up to multigigabits per second, since traditional third-generation(3G) and fourth-generation (4G) frequency bands are currently congested and might notbe sufficient to support the increasing quality of service (QoS) requirements in futurecellular networks. Broadband communications at high mobile user speed introducesadditional challenges to the modelling, design and analysis of 5G networks as it becomesdifficult to accurately estimate the rapidly varying channel and there is a significant limitto the coverage area and transmission rate.

In this thesis, we focus on transmission techniques for future wireless communica-tions in high mobility scenarios with HST as a case study. The HST is used as a casestudy due to the large scale and rapid deployment of high speed railway systems [6–8]. Adirect application of transmission techniques tailored for conventional cellular networksin high mobility scenarios is likely to be suboptimal because field test show that thedirect implementation of existing 4G systems in an HST network [9] can only providea data rate on the order of 2→ 4 Mbps. Hence, studies on how the communicationthroughput performance can be increased in a realistic high mobility scenario are carriedout in this study.

1.2 Challenges in high mobility scenarios

Providing broadband communication to users travelling at speeds up to 500 km/h witha data rate in the order of hundreds of Mbps or higher [10] is not a trivial task. Thepromising key elements/technologies to tackle traditional low mobility scenarios such asincreased bandwidth using HFBs [11], cell densification via reduced cell sizes [12],improved cell-edge coverage with multiple-input multiple-output (MIMO) relays [13]and increased spectral efficiency through advanced MIMO and interference coordinationtechniques [14] have to be addressed from a different angle when considering highmobility scenarios. This is as a result of the unique properties experienced in highmobility scenarios such as fast time varying fading channel due to the rapid change inenvironment, frequent handovers due to multiple cell crossings and the presence ofvehicular penetration loss (VPL), which can be in the range of 15-25 dB depending onthe operating frequency and vehicle type [15].

20

1.2.1 Fast time varying fading channels

One of the main characteristics of high mobility communications is the fast time-variation of the fading channel. The high speed of MT over a large distance causes a fastchange, both in the large-scale and short term fading of the channel, since the MT willlikely pass through different types of environment (e.g. tunnels, wide plains and hillyterrain) within a short time frame. In traditional systems, performance improvements canbe achieved as the MT observes the instantaneous channel and feeds the channel stateinformation (CSI) back to the base station (BS). The effect of the CSI feedback delaycan be significantly reduced using channel prediction schemes with the assumption thatthe channel exhibits temporal correlation. However, this assumption does not hold forhigh mobility scenarios, because the present and past CSI may be uncorrelated.

At high speeds, the effect of Doppler shift/spread is much more severe since, thechannel’s coherence time is inversely proportional to the Doppler shift, which dependson both the carrier frequency and the speed of the MT. For example, when an MTtravels at 500 km/h and the carrier frequency is 6 GHz, the maximum Doppler shiftis 2.78 kHz and the coherence time is approximately 152 µs, while at a speed of 30km/h the coherence time is approximately 2.5 ms. Thus, channel estimation becomesmore challenging at higher speeds because the estimation has to carried out within aperiod no longer than the coherence time (152 µs in the example). In reality, the speedof the MT can change over time and can change frequently, leading to time-varyingDoppler shifts/spread and non-stationary fading coefficients. The non-stationary effectand fast changing propagation environment makes accurate channel estimation difficult.Therefore, the assumption of perfect CSI adopted for the design of low mobility systemscannot be applied to high mobility systems.

1.2.2 Large and frequent resource handovers

A handover is the process by which an MT maintains an active connection whilemoving from one cell to another. When the MT is at the cell-edge, the MT searches forneighboring cells/BSs and sends a measurement report to the source cell/BS. Then, thesource BS selects the best neighboring BS using a set of criteria and hands over the MTto the best neighboring BS [16, 17]. The handover is usually triggered at the MT basedon two settings, "handover hysteresis" and "time to trigger" (TTT). The optimal valuesof these settings depend on different variables such as the MT speed, radio network

21

deployment, propagation conditions, and system load [18, 19]. An MT travelling athigh speed over an extended period of time will pass through multiple cells, therebytriggering frequent handovers. For example, an MT moving at 350 km/h, will initiatehandovers every 10-20 s for a cell size of 1-2 km. Hence, the handover procedure ismuch more challenging than in conventional systems with low mobility because theremight not be sufficient time for the system to complete the handover procedure.

Furthermore, large numbers of MTs might be moving at high speed at the same time,as large numbers of onboard users can be found in public transport vehicles. This resultsin a large number of handover process initiations, which requires a large amount ofnetwork resources. The small time-window to complete a handover process and thelarge number of handover processes initiated simultaneously will lead to excessivehandover failure if conventional handover schemes are directly implemented in highmobility scenarios.

1.2.3 Presence of vehicular penetration loss

With the increasing number of users accessing the wireless network at high mobility,most MTs will likely be in an enclosed vehicle. The materials used in the construction ofthe transport vehicle can significantly attenuate the received signal power in both uplinkand downlink. The loss in signal power, known as vehicular penetration loss (VPL), canbe quantified by the ratio of the received power immediately outside the vehicle to thereceived power inside the vehicle [20]. The value of VPL has been shown to depend onthe carrier frequency, the antenna position and the type of glass used for the vehiclewindows [21, 22]. In particular, HSTs are equipped with well-sealed train carriagesmade of special alloys that are difficult for signals to penetrate. The penetration lossesfrom train carriages are usually in the range of 20 to 35 dB for the 2 GHz frequencyband [8]. A large VPL will significantly degrade the BS-HST communication linkquality and make it more difficult to achieve reliable broadband communications forhigh mobility users.

1.3 Network design approach and assumptions for HSTcommunication

In this section, the network architecture used within the scope of this thesis is described.Assumptions made based on the network architecture and the rationale behind the

22

assumptions within the context of mobility management, channel estimation andfeedback delay are presented.

1.3.1 Network architecture

Currently, the one-hop network architecture, where there is a direct communicationlink between the BS and the onboard users in a train is mainly applied to the HSTnetwork. To mitigate the effect of frequent handovers between cells along the rail track,the use of enhanced cell combination has been proposed, where remote antenna units(RAUs) with the same frequency and parameter settings are used to enlarge the cell[23, 24]. However, due to the high VPL caused by the metalized windows and carriagesof the train, radio signals are received at significantly lower power. Hence, the use ofa two-hop network architecture is seen as a promising concept to provide improvedquality of service (QoS) to onboard users and to enhance cellular coverage. Varioustwo-hop network architectural schemes have been proposed for broadband wirelessaccess on HSTs [24–26], where the communication link is divided into BS-to-HST link(the backhaul link) and HST-to-onboard link (the access link).

M

Fig. 1. Cooperative MRN system model.

23

In most of the proposed two-hop network architectures, WiFi is employed on the accesslink inside the vehicle. However, the use of moving relay nodes (MRNs) on the accesslink, which is under the full control and management of the core network is mostlypreferable by mobile operators. In this thesis, we mainly adopt the use of relay systemsfor HST as shown in Fig. 1.

The use of MRN offers three main advantages. First, the relative positions ofonboard MTs in respect to the MRN changes little for the duration of the trip, ensuringgenerally good channel conditions that can support very high rate communications.Second, the signaling overhead is significantly reduced by congregating the handoverprocesses of large numbers of onboard MTs to an individual handover process of asingle MRN. Third, onboard MTs are relatively close to the MRN so that the transmitpower of onboard MTs can be reduced. Furthermore, with the increasing interest inthe use of mmWaves for 5G networks, highly directional beamforming antennas withelectronic steering capability can be potentially beneficial for the MRN backhaul link.Inband relaying, in which the access and backhaul links operate on the same frequencyband, can lead to access-to-backhaul and backhaul-to-access link interference. Thisinterference can be avoided with out-of-band relaying, where the access and backhaullinks operate on different frequency bands, but these comes at an additional cost to themobile operator. With the inband relaying, the interference can be significantly reduceddue to the high VPL of an HST, and can potentially be eliminated at higher operatingfrequencies.

The notion of the using relay systems for HSTs was first presented in [27]. Theidea is to equip an HST with multiple MRNs, which connects to the donor cellularnetwork(s) with multiple wireless backhaul links in a coordinated and controlled manner.This concept ensures that the access link is an integrated extension of the cellularnetwork and sufficiently high data rates are achieved due to the multiple backhaul linksestablished. The MRNs in each of the carriages are connected via the cooperative relayX2 (crX2) interface, which is assumed to be of zero latency and can be realized usingwired interface such as optical fiber or wireless interface operating outside the radioband frequency. Each of the MRNs has multiple antennas on the exterior and interior ofthe train and spans the length of each carriage.

24

1.3.2 Mobility management

Mobility management in an HST communication scenario faces the challenge of frequenthandovers, large signaling overheads due to large groups of MTs in the HST initiatingsimultaneous handovers and limited time for cell selection. Large simultaneous handoverprocesses and the high power consumption of the onboard MTs can be significantlyreduced with the adoption of the two-hop network architecture, since all the onboardMTs’ data is congregated at the MRN(s) which are equipped with external and internalantenna arrays. Therefore, handover processes are only initiated by the MRN(s), therebyeffectively eliminating the large signaling overheads due to the large group of MTs.

The deployment of radio remote units (RRUs) [24, 28] can be used to extendthe coverage of a single cell, thereby creating more time for a successful handoverprocedure/cell selection and reducing the number of handover failures. A dual linkhandover scheme was proposed in [29] to mitigate the effect of frequent handovers overthe backhaul link in HST communications. The proposed scheme adopts the use of tworelaying antennas mounted at the front and rear of the train, respectively. When thetrain moves towards the cell edge, the front antenna initiates the handover, while therear antenna maintains the connection to the serving BS. In [30], the problem of highhandover failure probability was addressed through the use of coordinated multiple-pointtransmission concept to achieve diversity gains with the HST receiving signals from twoadjacent BSs while moving through overlapping areas. Another approach is to exploitthe prior knowledge of the position of the HST and the deterministic route and directionof movement of the HST. In the context of the handover trigger in LTE/LTE-A, wherethe MT starts to measure the received signal strength (RSS) from all possible target BSsat the cell-edge and sends the measured report to the serving BS for further processing,the exploitation of prior information approach eliminates the search for target BSs andfeedback of measurement report. As soon as the BS establishes a connection withthe HST, the serving BS estimates the time the target BS should expect the HST. It ispossible to estimate the time due to prior knowledge of the position and speed of theHST, while the knowledge of the target BS is known as a result of the deterministic railroute and direction of movement of the HST. This approach significantly reduces thehandover delay and the potential target BSs along the rail route can reserve resources forthe admission of the MRN well ahead of time.

25

1.3.3 Channel estimation and feedback delay

Multiple-input multiple-output (MIMO) and orthogonal frequency division multiplexing(OFDM) schemes are considered to be reliable for providing high data rates [31] andare adopted mainly by the 4G wireless network and are most likely to be adopted bythe 5G wireless network. The performance of OFDM systems with high mobility islimited by the introduction of severe channel dispersions in the time and frequencydomains leading to inter-symbol-interference (ISI) and inter-carrier-interference (ICI).Thereby, these impose a great challenge on the accurate estimation of the channel.In general, techniques for channel estimation have been extensively studied [32–39]in the presence of fast time varying channels. Studies in [32, 33, 35], focus of blindestimation methods, where the need for training sequences is avoided and bandwidthis conserved in the process. To better track the channel variations, studies in [34, 36]proposed the cooperation of blind methods with part of the training data. Efficientchannel estimation techniques have been proposed for high mobility in [37, 38, 40–42],in which ICI mitigation is jointly taken into account. In particular, the linear minimummean square error (LMMSE) channel estimator in [37] has proved to be an effectivemethod for fast time-varying channel estimation in OFDM systems. Studies in [41],considered the ICI caused by large Doppler shifts in an HST communication system andshowed that the ICI can be mitigated by exploiting the train position information and thesparsity of the basis expansion model (BEM)- based channel model. When consideringhigher frequency bands (HFBs) for HST communication systems, there is an increasedsensitivity to the effect of Doppler shift. An efficient Doppler mitigation method basedon OFDM systems was proposed in [43], where milli-meter wave (mmWave) bands forHST communications were considered. Based on these studies, we make the assumptionin this thesis that the Doppler shift can be perfectly compensated for [44–46].

In practice, if the OFDM-based downlink transmission scheme of 3GPP-LTE/LTE-Ais considered, MTs are required to feedback a channel state report, which indicates theestimated instantaneous CSI to the BS. The BS uses the knowledge about the channelstate report to select a suitable modulation and coding scheme (MCS) in order to obtainan appropriate scheduling decision and data rate for transmission. However, in highmobility scenarios, the possibility of selecting an appropriate MCS, i.e., achievingoptimum link adaptation and channel dependent scheduling is limited because ofoutdated channel state feedback reports due to the delay between the point in time whenthe MT estimates the channel condition and the application of the channel state report

26

at the base station [47, 48]. Different methods have been proposed for handling thechallenges faced due to delayed feedback in wireless communication systems in highmobility scenarios. The present implementation to tackle the issue of feedback delay isto avoid link adaptation on the instantaneous channel, but to carry out link adaptation onthe long term average channel condition and rely on hybrid automatic repeat request(HARQ) with soft combining for rapid adaptation [49]. Authors in [50] proposed away of handling the delayed channel state report for very rapidly varying channels byincreasing the subcarrier bandwidth of the OFDM system while ignoring the impactof the cyclic prefix. The authors assume that the only effect of high mobility is anincrease in ICI, making the system more sensitive to fast fading channels. In [51], anadditional antenna was proposed to be placed in front of the transmission antennas inthe direction of travel for moving relay applications which would be used to predictthe channel condition. A different approach was considered in [52], specifically forthe HST rail environment, where the characteristics of the rail environment and theprior knowledge of the position and direction of movement of the HST are exploited.This method assumes the use of a two-hop network architecture and fixed externalantennas on the roof top of the HST with a relatively deterministic rapidly changingrail environment. Although these studies attempt to eliminate the effect of feedbackdelay, other problematic issues are introduced. Hence, the impact of feedback delay isconsidered in this thesis.

1.4 Literature review

This section provides a more detailed review of the existing and related literatureassociated with the scope of the thesis besides the challenges and existing workpresented in the previous section. The literature review related to transmission schemesfor HSTs, resource scheduling approaches for HSTs and higher frequency bands forHSTs are surveyed in Section 1.4.1, 1.4.2, and 1.4.3, respectively.

1.4.1 Transmission schemes for HSTs

The demand for cellular broadband wireless communications from high mobilityusers has tremendously increased due to the rapid deployment of large-scale publictransportation, such as HSTs, in many parts of the world. Providing high data rates andgood QoS required by onboard users in an HST is a challenging task in the presence of

27

rapidly varying channel conditions. An effective way to improve the data rate is to takeadvantage of the spatial dimension of the HST using a two-hop system architecture andapplying advanced MIMO and OFDM techniques [53].

MIMO spatial multiplexing in high speed railway communications has been studiedin [54–56]. Studies in [54] analyzed the achievable MIMO capacity for high speedtrains with multiple antennas based on 3-D modeling of the line-of-sight (LOS) channel.The authors show that spatial multiplexing gains can be achieved with a proposedmulti-group multi-antenna (MGMA) scheme, in which a capacity gain can be achievedby adjusting the weights among the MGMA arrays. On the other hand, the approachin [55], was to reconstruct the channel through the method of principal componentanalysis such that the correlation in the railway channel matrix is reduced. Hence,spatial multiplexing gains can be achieved. Reference [56] employs the selection of theappropriate antenna spacing to ensure that the channel response is not rank deficient.The impact of imperfect CSI on high mobility systems, examined in [57], shows that theaccurate estimation and tracking of the fast time-varying fading is critical to reliableoperation. Studies in [58] have shown that in many mobile scenarios, the upper boundof the achievable data rate is mainly defined by the feedback delay of the CSI. Dueto the outdated and imperfect CSI, the throughput gains provided by conventionalprecoding methods may be marginal or even non-existent. Conceptually simple antennaselection methods implemented at the transmitter have been studied with differentreceive antenna combinations and transmission schemes [59–63]. The performance oftransmit antenna schemes relies on the required feedback of information on selectedantenna index/indices from the receiver to the BS. However, most of these studiesonly consider single antenna selection. In [61], multiple transmit antenna selection forspatial multiplexing was achieved by considering the detection order of an orderedsuccessive interference cancellation (SIC) receiver with power control per antenna.Antenna selection for spatial multiplexing for practically realizable receivers was studiedin [62] and [63]. The number of transmitted streams in these studies were fixed and asubset among all possible transmit antennas was chosen.

The aforementioned antenna selection schemes are designed from the perspective ofconventional cellular systems. However, there are two major differences between anHST system (considering the BS to train backhaul link) and a conventional cellularsystem. The number of receiving antennas and the speed of the receiver can both benotably higher in the HST scenario. These characteristics suggest that large antennaarrays can be used to improve the throughput performance of an HST communication

28

system via the achieved multiplexing gain. Moreover, the size of the train can beexploited to ease the spatial separability of the transmitted signals on the receiver side.Due to the high speed of the train, conventional precoding methods may not provide anypractical performance gains. Instead, simpler spatial multiplexing methods with antennaselection schemes could be of practical interest. However, with large antenna arrays,adaptive transmission schemes may require extensive computational load, especially ifthe design is based on the exhaustive search type of adaptivity. Practical alternativeschemes to the exhaustive search type of adaptivity were proposed in the author’scontribution in [64].

1.4.2 Resource scheduling approach for HSTs

One advantage of the OFDM technique is the ability to split the system bandwidth intoseveral orthogonal channels, thereby providing the possibility of sharing radio resourceswith multiple users within the same time slot. Maximizing the downlink throughputof moving relay nodes installed on an HST within an LTE-A cellular communicationsystem was studied in [64–70]. However, the issues of fairness and the impact onthe throughput of macro users scheduled along with the MRNs were not addressed.Resource scheduling approach in HST wireless communication networks is one of thekey issues for improving the resource utilization in the entire network. With the rapiddeployment of HSTs around the world, it is of interest to develop a scheduling schemethat provides a trade-off between fairness and throughput for different types of users(low and high mobility users). Various scheduling strategies have been proposed fortraditional OFDM systems. For example, studies in [71] demonstrate that enhancedsystem throughput and fairness among users can be achieved with maximum carrierto interference ratio scheduling. Proportional fair (PF) scheduling was proposed forOFDM systems in [72–74] where fairness among users was addressed while maximizingsystem throughput. In one of the schemes proposed in [72], users are allocated resourcesacross multiple subchannels in the same time slot without updating the average datarate until the end of the time slot. A matrix-based calculation was used to updatethe average data rate in the PF ratio in [73]. In this approach, each user is assignedto a subchannel and a number of time slots are allocated to each user based on theaverage data rate. A dynamic resource allocation scheme for downlink MIMO-OFDMsystem was proposed in [75] which exploits all the spatial subchannels to provideimproved multiuser diversity gains. The proposed scheme aims to provide a good

29

trade-off between performance and complexity without considering fairness betweenusers. A user priority resource allocation scheme for an LTE-A downlink system wasproposed in [76]. The proposed scheme uses a fuzzy inference system to effectivelydetermine a suitable priority value for each user based on the reported channel statereport and QoS fulfilment information (QFI), so as to enhance system throughput andattain a high degree of fairness. With the introduction of relaying concept into thecellular network, MTs, particularly cell-edge MTs, can achieve higher data rates usingmultihop links through fixed relay nodes. Some studies have addressed the issue ofscheduling in the presence of relays, such as in [77–79]. A cost aware PF schedulingscheme was proposed for cellular networks with relays in [77], where the schedulingdecisions are made at the BS, but the scheduling algorithm runs at both the BS and therelay. In [78], a centralized downlink scheduling scheme in a cellular network with asmall number of relay nodes was proposed. The aim of the proposed scheme was toguarantee the stability of the final destination user queues for the largest set of arrivalrates and the effect of the number of relays used. In [79], three different schedulingalgorithms were compared in a relay enhanced cellular network. The proposed partialPF scheme employs a two-dimensional PF algorithm for the resource allocation ofthe second sub-slot and assigns the resource units of the first sub-slot according to theprevious allocation result. In [80], a resource allocation scheme for OFDM systems inHST scenarios was developed in which multiple bit error rate (BER) requirements fordifferent types of packets in one data stream were considered. The scheme also tookinto account the channel estimation error caused by the Doppler shift by grouping anumber of contiguous subcarriers into chunks and allocating resources on a chunk bychunk basis. A resource allocation optimization problem in HST wireless networks wasformulated in [81], in which subcarriers, antennas, time slots and power are jointlyconsidered with a solution obtained by quantum-behaved particle swarm optimization.In all of these aforementioned investigations, none of the studies take into account theimpact of a scheduling approach between different groups of mobile terminals in termsof fairness and throughput performance for each group of mobile terminals. The impactof scheduling considering different groups of mobile terminals was addressed by theauthor’s contribution in [82].

30

1.4.3 Higher frequency bands for HSTs

Different transmission schemes [64] and techniques [54] have been considered tomeet the rapid increase in QoS demand on HSTs due to the evolution of railwaycommunication services, such as onboard high definition (HD) video surveillance,onboard high real-time data rates, real-time train dispatch HD video, and train operationinformation [83]. However, these schemes are limited by the scarce available bandwidthat microwave frequency bands, since the additional railway communication servicesrequire huge amounts of bandwidth, up to tens of GHz [84]. Large available bandwidthcan be found at higher operating frequencies and these available spectra can be upto 200 times more than at microwave frequency bands. Therefore, a drastic increasein capacity of the HST communication network is potentially possible if HFBs areconsidered. With the large available bandwidth at HFBs seen as a potential way to meetthe very high and increasing data rate demand, the use of HFB has gained growinginterest in the development of 5G wireless communication networks [85, 86]. However,due to the sensitivity to the atmosphere and severe propagation loss experienced at thesefrequency bands, long range transmission is challenging. On the other hand, with thesmall wavelength and recent advances in modular antenna array technology for HFBs[87], high beamforming gains can be obtained to combat large propagation losses andtherefore improve the suitability for outdoor communication.

The limitation of microwave bands for future railways was highlighted in [84]and the use of HFBs was suggested for future 5G communication system for railways(5G-R). In [88], a hybrid spatial modulation beamforming scheme was proposed atHFBs under the HST scenario, where a combination of spatial modulation and hybridbeamforming is used to enhance rate performance. However, channel state informationfeedback delays were not taken into account, which can significantly affect the rateperformance. A modification to the IEEE 802.11ad beam sweeping approach wasexamined in [89], where the number of beams and an optimum repetition time to sweepthrough the beams was determined from the velocity estimate of the HST. Additionally,HFB beam switching support for HSTs was considered in [90], in which the beamswitching approach leverages the knowledge of the train position for optimizing thebeamwidth to achieve a higher rate. However, information about the selected beamat the BS is required at the HST and the beam design details were not examined. AGrey Markov chain method was used to eliminate the residual error caused by thehigh mobility of the HST between the estimated angle of arrival (AoA) and angle of

31

departure (AoD) in [91] using past and present predicted values. The Grey Markov chainbased beamforming algorithm uses the Grey theory component to reflect the strongregularity of the HST and the Markov chain component reflects the irregular factors,such as abrupt speed change moments. The authors assumed a perfect estimation of theuplink AoA’s of the past using multiple signal classification (MUSIC) or estimationof signal parameters via rotational invariance techniques (ESPRIT), which are AoAestimation methods that are not entirely suitable for high speed scenarios. In the author’scontribution in [92] a timer-based beamforming selection scheme was proposed wherethe impact of feedback delay is avoided, and the AoA/AoD is proactively and easilyestimated, since LOS propagation is assumed.

1.5 Objectives, contribution, and outline of the thesis

The aim of this thesis is to develop algorithms that maximize the achievable throughputfor high mobility users on public transport vehicles in future wireless communicationsystems. Specifically, emphasis is placed on HSTs with a two-hop network architectureas shown in Fig. 1. Parts of the thesis are based on system level simulations andoptimization theory. The thesis focuses on the performance improvement in thethroughput of onboard users in an HST through the use of multiple MRNs equipped onthe HST. The onboard users are assumed to be in relatively low mobility and in closeproximity with respect to the MRNs. Hence, attention is given to the backhaul link,which is considered the main bottleneck in the two-hop network architecture. Thisthesis also distinguishes the difference between the traditional cellular network and thecharacteristics in an HST network structure. The designs and techniques developed inthis thesis leverage the peculiar characteristics of the HST network. In the following, theoutline of this thesis is provided giving the main contributions contained in each chapter.Chapter 2 is mainly based on [64]. This chapter focuses on maximizing the downlinkthroughput of the BS-to-train link. The analysis is built on MIMO-OFDM techniquesfor theoretical link-level and practical system-level scenarios. First, a throughputmaximization problem in a theoretical single-cell MIMO-OFDM train scenario isconsidered and two convex-optimization-based transmission strategies are described.Then, practical low-complexity MIMO-OFDM transmission schemes based on simpleantenna selection (AS) methods with spatial multiplexing are proposed. Practicalsystem-level performance and complexity evaluation of the proposed transmissionschemes are compared with LTE based open-loop and closed-loop precoding methods.

32

Chapter 3 relies on results documented in [82, 93]. A system model for the downlinkcellular network with an HST equipped with an MRN on each carriage of the HST ismodelled. The radio resources available in each cell are shared optimally and fairlyamong MTs within the cell. However, the MTs are grouped into two types, regular lowmobility users a.k.a. ground macro users (GMUs) and MRNs when the HST is presentin the cell. The effect of joint and disjoint scheduling approaches w.r.t. the types ofusers are analyzed. Two hybrid scheduling algorithms are proposed and applied withinthe LTE framework. The amount of resources assigned to the MRNs will depend on thenumber of active onboard users (OBUs) associated with each MRN. Hence, the effect ofreducing the number of MRNs on an HST is analyzed with the assumption that theMRNs can cooperate to serve OBUs in carriages with no MRN.Chapter 4 is based on results detailed in [92, 94]. The impact and feasibility of theuse of higher frequency bands for transmission on HST communication network wasanalyzed. A modification to the OFDM frame structure was proposed based on agenerated lower bound SNR/SINR to achieve successful transmission. A sequentiallyordered codebook for HST communications was developed, where the codebook relieson the array response vector with ordered angular inputs generated from a range ofpossible AoA/AoD values. A distance/time-based beamforming selection scheme forHST communications was proposed, where the scheme leverages the LOS propagationand prior knowledge of the HST position and velocity.Chapter 5 concludes the thesis and provides possible future directions based on thiswork.

1.6 The author’s contribution to the publications

The author of this thesis has contributed to seven published papers [64, 66, 67, 82, 92–94] and one patent [52]. This thesis is based in parts, on two journals [64, 94] andthree conference papers [82, 92, 93]. All journals and conference papers have beenpublished. The author of this thesis had the main responsibility in conceiving the originalideas, deriving the algorithms, implementing the algorithms on link and system levelsimulators based on MATLAB, generating the numerical results, and writing the papers.Other authors provided comments, criticism, and support during the process.

33

34

2 Transmission strategies for throughputmaximization in HST communications

This chapter considers both theoretical and practical transmission schemes for high speedtrain (HST) communication networks. The key performance objective is throughputmaximization with practical system level performance and complexity evaluations.In Section 2.1, the system models for both theoretical and practical HST scenariosare introduced. The throughput maximization algorithms for both cooperative andnon-cooperative HST scenarios are described for the theoretical HST system in Section2.2. State of the art transmission strategies as applied to HST communication networksand the proposed low-complexity antenna selection (AS) and spatial multiplexingmethods are described in Section 2.3. The performance of the proposed algorithms isevaluated in Section 2.4 and the chapter is summarized in Section 2.5.

2.1 System model

A train communication scenario is considered with a focus on the BS-to-HST communi-cation link. The train has multiple carriages, each equipped with a single moving relaynode (MRN). The number of MRNs on the HST is equal to the number of carriages andis denoted by M. Each MRN has an external antenna array with Nr receive antennas thatare assumed to be evenly spaced along the length of the corresponding carriage. TheMRNs are also equipped with internal antennas to serve onboard users. The number oftransmit antennas at the BS is denoted by Nt and the MIMO-OFDM communicationscheme is adopted with the frequency domain divided into C subcarriers. The downlinkcommunications between the BS and the train is analyzed in simple single-cell andrealistic multi-cell systems to reflect theoretical and practical scenarios, respectively.For the clarity of presentation, separate signal models are presented for both scenarios.In all cases, we assume linear transmit/receive design strategies.

2.1.1 Theoretical single-cell scenario

In the theoretical single cell scenario, two modes of operation are derived. These arenon-cooperative and cooperative mode of operation. In the non-cooperative mode, each

35

MRN performs receives independently and there is no form of cooperation between theMRNs. In the cooperative mode, the MRNs can cooperate with each other and act as alarge receive antenna array in order to aid data reception. The aim of the theoreticalsingle-cell scenario is to provide fundamental insight into the problem rather thanproviding practically realizable algorithms.

MRN non-cooperative mode

With a BS connected to the M MRNs, the BS transmits data to all MRNs simultaneouslywhile each MRN performs reception independently at the train. This is illustrated inFig. 2 without the connecting link between the MRNs. This scenario is conceptuallysimilar to a point-to-multipoint MIMO-OFDM system. The downlink received signalvector yc,m ∈ CNr at the mth MRN for the cth subcarrier is given by

yc,m =M

∑m=1

Lm

∑l=1

Hc,mmc,m,lsc,m,l +nc,m (1)

where Hc,m ∈ CNr×Nt is the channel matrix between the BS and the mth MRN, mc,m,l ∈CNt is the unnormalized precoding vector for the lth stream of the mth MRN, sc,m,l ∈ Cdenotes the corresponding data symbol and nc,m ∼ CN (0,N0INr) is the additive complexwhite Gaussian noise vector with zero mean and N0 variance per element. The numberof data streams given to the mth MRN is denoted by Lm. The total transmit power of theBS is given by P = ∑

Cc=1 ∑

Mm=1 ∑

Lml=1 ‖mc,m,l‖2

2. The received signal to interference plusnoise ratio (SINR) of the lth stream at the mth MRN can be written as

Γc,m,l =|wH

c,m,lHc,mmc,m,l |2

‖wc,m,l‖2N0 +M

∑q=1

Lq

∑i=1

(i,q)6=(l,m)

|wHc,m,lHc,mmc,q,i|2

,(2)

where wc,m,l ∈ CNr denotes the receive filter for the lth stream of the mth MRN.

MRN cooperative mode

The BS serves the whole train simultaneously and all MRNs can jointly performreception, as depicted in Fig. 2. Due to cooperation between the MRNs, this scenariocan be conceptually interpreted as a point-to-point MIMO-OFDM system.

36

MRN 1 MRN 2

BS

MRN M

Fig. 2. Single-cell train scenario [64] ©2016, IEEE.

The signal model is simplified since all MRNs can be considered as a single receiverand the index of the MRNs can be omitted.

The downlink received signal vector yc ∈CMNr for the cth subcarrier can be expressedas

yc =L

∑l=1

Hcmc,lsc,l +nc(3)

where mc,l ∈ CNt denotes the precoding vector of the lth stream, sc,l ∈ C is the corre-sponding data symbol and nc ∈ CMNr is the additive complex white Gaussian noisevector. The total number of data streams transmitted to all M MRNs is given byL ≤ min(Nt ,MNr). The channel matrix from the BS to the whole MRN system isdenoted by Hc = [HH

c,1 . . .HHc,m . . .HH

c,M]H ∈ CMNr×Nt . The total transmit power is ex-pressed as P = ∑

Cc=1 ∑

Ll=1 ‖mc,l‖2

2. Given the receive filter wc,l ∈ CMNr over all MRNs,the SINR for the cth subcarrier on the lth stream can be expressed as

Γc,l =|wH

c,lHcmc,l |2

‖wc,l‖2N0 +L

∑i=1i6=l

|wHc,lHcmc,i|2

. (4)

37

2.1.2 Practical multi-cell scenario

A schematic representation of the HST in a multi-cell network configuration is shown inFig. 3. The system consists of trisector antenna sites, where each sector forms a cell. Forthe ease of presentation, we refer to sectors as BSs in the rest of this chapter. Each MRNcan only have a direct connection to a single BS at a time. In other words, a single MRNcannot be served by multiple BSs simultaneously.

MR

N2

MR

N8

MRN 1 MRN 2

BS

X2

BS

Trisector BS

Ground UsersMRN M

Fig. 3. Multi-cell train scenario [64] ©2016, IEEE.

Different MRNs of the same HST can be connected either to the same BS or differentBSs. Thus, an HST with multiple MRNs can be served by a single BS or multiple BSs,depending on the position of the HST in the network layout. The number of MRNsjointly associated with BS b is denoted by Mb. Interference from onboard users to theMRNs’ external antennas is omitted since half duplexing operation is assumed. Weassume perfect synchronization since frequency offsets caused by the velocity of theHST can be compensated for at different positions along the predictable route of thetrain despite the fast time varying Doppler shift [44, 45]. We also assume that Q groundmacro users with low mobility are randomly distributed in the multi-cell system and aretherefore scheduled alongside the MRNs. Transmissions to ground users are considered

38

as interference. We model the fast fading channel based on a geometry-based stochasticmodeling approach, which allows the creation of an arbitrary double direction radiochannel independent of any antenna configuration. Other channel parameters are basedon statistical distributions extracted from real channel measurements and are obtainedstochastically [95]. More details on the used channel model are given in Section 2.4.1.

In this scenario, we consider an MIMO-OFDM system operating in frequencydivision duplexing (FDD) mode and assume that the MRNs served by the same BScan cooperate. The number of links formed depends on the number of BSs associatedwith the HST at each time instance. The MRNs coordinate between themselves andform groups based on the MRNs associating with the same BS and each group formedcooperates as a single receive antenna array to establish a backhaul link. For clarity,the whole train is assumed to be served by a single BS in the following signal modelrepresentation (i.e., Mb = M). The received signal vector yc ∈ CMNr for the MIMOtransmission for subcarrier c can be expressed as

yc = HcMcsc +K

∑k=1

Hc,kMc,k sc,k +nc (5)

where Hc ∈ CMNr×Nt is the channel matrix between the serving BS and the entireMRN system, Mc = [mc,1 . . .mc,L] ∈ CNt×L is the unnormalized transmit precodingmatrix for the desired data streams, sc ∈ CL is the corresponding transmit signal vectorand nc ∈ CMNr is the additive white Gaussian noise vector. Note that Mc can beexpressed as Mc = P1/2

c Fc, where Fc = [fc,1 . . . fc,L] is the normalized precoding matrixand Pc = diag(Pc,1 . . .Pc,L) consists of the powers allocated to each of the L streams.Hc,k ∈ CMNr×Nt , Mc,k = [mc,1 . . .mc,Lk

] ∈ CNt×Lk , and sc,k ∈ CLk are the channel matrixfrom interfering BS k to the MRN system, the precoder matrix used at BS k to transmitLk (interfering) data streams and the corresponding signal vector. The total number ofthe interfering BSs is denoted by K.

The throughput optimal linear minimum mean square error (MMSE) filtering isapplied at the receiver. The MMSE filter Wc = [wc,1 . . . wc,L] ∈ CMNr×L is obtained byminimizing the mean square error defined as E

[‖sc−WH

c yc‖2], where

Wc = (HcMcMHc HH

c + Rc)−1HcMc , (6)

where Hc and Rc are the estimates of the channel matrix Hc and the interference plusnoise covariance matrix Rc, respectively. The matrix Rc is given by

Rc =K

∑k=1

Hc,kMc,kMHc,kHH

c,k +N0IMNr . (7)

39

The SINR of transmission stream l for subcarrier c at the output of the MMSE receiveris given by

Γc,l =|wH

c,lHcmc,l |2L∑

j=1, j 6=l|wH

c,lHcmc, j|2 + wHc,lRcwc,l

. (8)

2.2 Transmission techniques from theoretical perspective

In this section, the throughput maximization problem is formulated for a downlinksingle-cell MIMO-OFDM train system with and without cooperation between MRNs.It is assumed that the channel is known at the BS and the MRNs, since the impact ofchannel estimation error and feedback delay on both the MRN cooperative mode andMRN non-cooperative mode are assumed to be the same. Hence, we mainly investigatethe performance between the MRN cooperative mode and MRN non-cooperative modein the presence of different channel characteristics and propose precoding schemes thatmaximize the throughput given the aforementioned assumptions.

2.2.1 MRN non-cooperative mode

In order to maximize throughput, the optimization problem of sum rate maximizationunder total transmit power constraint can be formulated as follows

maximizemc,m,l ,wc,m,lC,M,L

c=1,m=1,l=1

C

∑c=1

M

∑m=1

L

∑l=1

log(1+Γc,m,l

)subject to

C

∑c=1

M

∑m=1

L

∑l=1‖mc,m,l‖2

2 ≤ P.

(9)

Note that a positive weighting factor can be added to the objective function of (9) inorder to maintain a certain degree of fairness among the MRNs, and possibly reflectthe number of onboard users in each carriage required to be served. Problem (9) isnon-convex, and thus, it cannot be solved in its current form. However, (9) can bereformulated and approximated to obtain an efficient (sub-optimal) solution where theobjective function (i.e., the sum rate) converges. However, global optimality cannot beguaranteed due to the non-convexity of the original problem. The proposed algorithm isan extension of the approach in [96] to a multi-carrier system. The idea of the algorithmis to divide (9) into precoder and receive filter design problems, which are solved

40

alternately until the objective function converges. In other words, the precoders areoptimized while the receive filters are fixed, and vice versa. The optimal receive filtersper-subcarrier are obtained by using the linear MMSE criterion. This is given by

wc,m,l =

(M

∑m=1

L

∑l=1

Hc,mmc,m,lmHc,m,lH

Hc,m + INr N0

)−1

Hc,mmc,m,l . (10)

In order to solve the precoders, the sum rate maximization problem can be reformulatedas a sum log-mean square error (MSE) minimization problem, expressed as

minimizemc,m,lC,M,L

c=1,m=1,l=1

C

∑c=1

M

∑m=1

L

∑l=1

log(εc,m,l

)subject to

C

∑c=1

M

∑m=1

L

∑l=1‖mc,m,l‖2

2 ≤ P

(11)

where εc,m,l is the MSE at the mth MRN for the cth subcarrier on the lth stream and therelation between the MSE and SINR [97] is expressed as

ε−1c,m,l = 1+Γc,m,l . (12)

With fixed receive filters, the sum log-MSE minimization problem is still a non-convexproblem. Therefore, the transmit precoder design is reformulated using difference ofconvex function program (DCP) by introducing an auxiliary constraint

εc,m,l ≤ 2−tc,m,l (13)

where the auxiliary variable tc,m,l is assumed to be such that 2tc,m,l ≥ 1. This impliesthat the domain of the convex MSE upper bounding function 2−tc,m,l is in the range ofpossible MSE values, i.e., between 0 and 1. Note that also other exponential functionsare applicable if they satisfy these assumptions. Applying (13) and relaxing the objectiveaccordingly, the resulting problem is given by

maximizemc,m,l ,tc,m,lC,M,L

c=1,m=1,l=1

C

∑c=1

M

∑m=1

L

∑l=1

tc,m,l

subject toC

∑c=1

M

∑m=1

L

∑l=1‖mc,m,l‖2

2 ≤ P

εc,m,l ≤ 2−tc,m,l ∀c,m, l.

(14)

41

The relaxation is tight since the inequality constraints achieve the optimal solutionwith equality. Problem (14) is still non-convex. Next, we apply successive convexapproximation (SCA) method to approximate (14), and then, we iteratively solve theresulting convex problems so that the objective function converges. Now, a linearapproximation of f (tc,m,l) = 2−tc,m,l is found by taking the first order Taylor seriesapproximation at a point t(i)c,m,l . This is expressed as

f (tc,m,l , t(i)c,m,l) = f (t(i)c,m,l)+(tc,m,l− t(i)c,m,l) f ′tc,m,l

(t(i)c,m,l)

=−a(i)c,m,ltc,m,l +b(i)c,m,l

(15)

where f ′ denotes the partial derivative and

a(i)c,m,l = ln(2)2−t(i)c,m,l , b(i)c,m,l = 2−t(i)c,m,l(

1+ ln(2)t(i)c,m,l

). (16)

As a result, we can formulate the following optimization problem at the ith iteration ofthe SCA method for fixed t(i)c,m,l

C,M,Lc=1,m=1,l=1

maximizemc,m,l ,tc,m,lC,M,L

c=1,m=1,l=1

C

∑c=1

M

∑m=1

L

∑l=1

tc,m,l

subject toC

∑c=1

M

∑m=1

L

∑l=1‖mc,m,l‖2

2 ≤ P

εc,m,l ≤−a(i)c,m,ltc,m,l +b(i)c,m,l .

(17)

The next point of approximation t(i+1)c,m,l can be found by using a line search method

in [96]. The SCA-based convex problems with the updated approximation points arerepeatedly solved. The objective value converges since it increases monotonically ateach iteration of the SCA method and the receive filter update via the MMSE. Thecomplete method is described in Algorithm 1.

2.2.2 MRN cooperative mode

To maximize throughput, well known singular value decomposition (SVD) and water-filling principles can be exploited. Specifically, we apply SVD to each per-subcarrierchannel matrix and use the resulting right singular vectors as precoders and left singularvectors as receive filters. This is written mathematically as Hc = UcΛcVH

c , where thecolumns of Vc ∈ CNt×Nt and Uc ∈ CMNr×MNr are the right and left singular vectors.

42

Algorithm 1 Precoder and Receive Filter Design for MRN Non-cooperative mode.Initialization:

1: i← 02: Initialize precoders and point of approximations mc,m,l , t

(0)c,m,l

C,M,Lc=1,m=1,l=1

Iteration:3: repeat4: Compute receive filters wc,m,lC,M,L

c=1,m=1,l=1 using (10)

5: repeat6: Compute SCA values a(i)c,m,l ,b

(i)c,m,l

C,M,Lc=1,m=1,l=1 using (16)

7: Compute precoders mc,m,lC,M,Lc=1,m=1,l=1 using (17)

8: Update tc,m,l using line search method9: i← i+1

10: until desired level of convergence.11: until desired level of convergence.

The diagonal entries of Λc ∈ RMNr×Nt are the ordered singular values (λc,1, ...,λc,L) ofHc with the remaining entries equal to zero. The optimal power allocation is achieved byextending the MIMO water-filling algorithm in [98] to MIMO-OFDM systems, wherethe resulting water-filling algorithm is performed over the independent MIMO channelsand subcarriers. This transmission technique is also capacity achieving strategy. Theoverall transmitter-receiver strategy achieves the channel capacity as follows

C =C

∑c=1

L

∑l=1

log

(1+

P∗c,l |uHc,lHcvc,l |2

N0

), with

P∗c,l =

(µ− N0

|uHc,lHcvc,l |2

)+ (18)

where the so-called water-level µ is chosen to satisfy the total power constraint,i.e., ∑

Cc=1 ∑

Ll=1 P∗c,l = P. Symbols vc,l and uc,l denote the lth column of Vc and Uc,

respectively.

2.3 Transmission techniques from practical perspective

In this section, details of the state of the art transmission scheme as applied to HST i.e.,LTE precoding scheme is first presented. Then two practical transmission algorithms are

43

developed to improve the throughput performance in an HST communication scenario.The proposed algorithms are based on antenna selection and spatial multiplexing.

2.3.1 LTE precoding schemes applied to high speed train scenario

This section examines the existing LTE precoding codebook schemes and how weapply them to the HST scenario for performance evaluation. The type of the codebooksused and the rationale behind the choice is briefly explained first. Precoding at the BSrequires information about the MIMO channel. The channel information can be obtainedvia feedback from the MRNs. Unfortunately, the feedback of the channel informationinvolves a significant overhead on the uplink capacity for most systems, especially onsystems with high bandwidth and/or high mobility [99]. Thus, a necessary and practicalsolution is the use of limited feedback, where both the BS and radio terminal share acommon codebook. The radio terminal searches the codebook for vectors/matrices thatcan maximize the overall system performance and the index of that vector/matrix is fedback to the BS. A combination of discrete Fourier transform (DFT) based codebook[100] and householder (HH) based codebook [101] is adopted by LTE and LTE-A dueto the high precoding gain, lower feedback overhead, lower complexity and flexiblesupport for various antenna configurations.

Based on the HH codebook design, we examine the impact of the LTE basedclosed-loop and open-loop spatial multiplexing schemes on the railway network. Inthe closed-loop spatial multiplexing scheme, the MRN/receiver estimates the channelfeedback information, which includes the channel quality indicator (CQI), precodingmatrix indicator (PMI), and rank indicator (RI) and feeds back the channel feedbackinformation to the BS in order to maximize the spectral efficiency. The CQI value, whichis based on the derived SINR, indicates the optimum modulation and coding scheme(MCS) to be used for the next transmission. The PMI chooses the optimum precodermatrix from a predefined codebook and the index of the chosen precoder matrix isconveyed to the BS, while the RI chooses the optimum number of layers for MIMOtransmissions. In the case of the open-loop spatial multiplexing scheme, the estimatedchannel feedback information includes only the CQI and the RI. The feedback of thePMI is not required, however, a subset of the codebook is applied at the BS togetherwith cyclic delay diversity (CDD). The open-loop pre-coding matrix is defined as

Fc = Fn[i]DcU (19)

44

where Fn[i] ∈ CNt×L is the ith precoding matrix from the chosen subset of the nthcodebook, Dc is a CDD matrix of size L×L that changes with the index of the sub-carriers, U is a fixed DFT matrix of size L×L. The CDD Dc and the DFT U matricesare defined for 2, 3, and 4 transmission layers in [102], [103].

The selection criteria for the PMI and RI are based on maximizing the throughput,which is defined as

Tr, f =C

∑c=1

r

∑l=1

log2(1+Γc,l

)(20)

where the subscript f signifies the selected precoder matrix and r, which can vary from1 to ≤min(Nt ,Nr), is the number of transmission rank/streams selected. The SINRΓc,l is taken from (8). When fixed rank transmissions are considered r = min(Nt ,Nr).The throughput maximizing PMI and RI values for the closed-loop scheme involve anexhaustive search through the codebook, which grows in size exponentially with anincrease in the number of transmit antennas. These values are chosen according to

(r∗, f ∗) = arg maxr∈R , f∈P

Tr, f (21)

where r∗ is the selected transmission rank from the set R , which consists of possibletransmission ranks, i.e, R = 1, ...,min(Nt ,Nr). The index f ∗ is the chosen precodermatrix from the codebook P , which includes the predefined set of precoders for differenttransmission ranks. On the other hand, the throughput maximizing RI value in theopen-loop spatial multiplexing scheme is chosen according to

r∗ = argmaxr∈R

Tr, f (22)

where f is a precoder matrix selected in a predefined and deterministic way accordingto (19). It is worth noting that even if the transmission rank changes, all the transmitantennas are always used for data transmissions.

2.3.2 Practical transmission schemes based on antenna selectionand spatial multiplexing

In this section, two low complexity transmission schemes with simplified antennaselection and spatial multiplexing are proposed for the HST scenario. Before a detaileddescription, the rationale behind these schemes is given. There are two fundamentaldifferences between an HST system (considering the BS to train backhaul link) and a

45

conventional cellular system. The speed of the receiver and the number of receivingantennas can be both significantly higher in HST scenarios than in the traditional cellularsystem. Thus, designing practical transmission schemes for the HST scenario requiresa change of perspective to start with. In LTE systems, the use of codebook-basedprecoding schemes can provide improved throughput performance in conventionalcellular scenarios. However, in high mobility scenarios, such as the HST scenario, theachievable gains in throughput performance by using precoding may be minimal or evennon-existent compared to transmission strategies without precoding, as will be shownvia numerical examples in Section 2.4.2. Moreover, conventional LTE-based dynamicrank transmissions, where the number of data streams is adapted to the prevailingchannel conditions, are mainly designed for mobile users with a low number of receivingantennas. Using LTE-based precoding schemes with dynamic rank transmissions, theperformance metric is calculated for all different precoder/stream combinations, and theone with the highest value is selected for the next transmission. This exhaustive type ofsearch process may require an extensive amount of computation since the number ofdifferent combinations can be high, especially with large antenna arrays also at thereceiver side.

In this respect, we propose two low complexity algorithms that are based on spatialmultiplexing and simplified antenna selection. The algorithms are less complex andprovide improved performance compared to the LTE precoding schemes with dynamicrank transmissions. Furthermore, these low complexity algorithms are also applicableto the case where large number of antennas are used at the transmitter and receiver.Computation is performed at the receiver side of a communication link by utilizingthe estimated MIMO channels using cooperative MRNs on the train. The informationon the selected number and positions of the transmit antennas is fed back to the BS,which then performs data transmission using spatial multiplexing with the chosen set ofantennas. In the spatial multiplexing technique used, precoding is not employed, andeach data stream is sent from a single transmit antenna. At the receiver side, each MRNperforms the linear MMSE reception. Due to cooperation between the MRNs, all theMRNs connected to the same BS within a transmission time interval can be seen as alarge MMSE receiver.

46

Adaptive transmission scheme based on two-phase antenna selection

The first practical algorithm has two phases for a proper selection of the transmitantennas. The target of the first phase is to select an appropriate number of transmitantennas to be used for the spatial multiplexing transmission. The second phase aimsto select the proper antenna positions for the given number of antennas chosen in thefirst phase. In both phases, the appropriate options are chosen that provide the highestestimated throughput assuming spatial multiplexing transmission. There are differentways to estimate the throughput performance at the receiver side. Since the MMSEreceiver is used, it is logical to calculate post-MMSE throughput by assuming spatialmultiplexing transmission, as given by

TNt =C

∑c=1

Nt

∑l=1

log2

(1+

Pc,l ||wHc,l hc,l ||2

N0 +Nt

∑i6=l

Pc,i||wHc,ihc,i||2

), (23)

where wc,l is the MMSE receiver for stream l at subcarrier c, and it is obtainedfrom (6). For later usage, the MIMO channel at subcarrier c is denoted by Hc =

[hc,1, ..., hc,Nt ]. Equation (23) gives the total throughput over the subcarriers andtransmission streams. The index Nt denotes the number of transmit antennas used.Another possible performance metric is pre-MMSE throughput given by

TNt =C

∑c=1

Nt

∑l=1

log2

(1+

Pc,l ||hc,l ||2

N0 +Nt

∑i6=l

Pc,i||hc,i||2

). (24)

Note that (23) is more accurate metric, whereas (24) is computationally less complex. Itis worthwhile mentioning that alternative performance metrics than throughput can alsobe used. One such metric can be composed by calculating the received signal powersand correlation levels for different combinations of antenna counts and positions. Thisidea is described in [66].

The key idea of the algorithm is to significantly reduce the number of combinationswhich need to be compared to each other. This is obtained by selecting only a subset ofpossible combinations to be compared with each other in both phases of the algorithm.The intuition here is that the most relevant combinations are chosen for the subsetsto be compared with each other, and the less relevant ones are left out from thecomparison. The sizes of the subsets can be considered as design parameters, and there

47

is a complexity-performance trade-off. Here the complexity refers to computational load,which is defined as the number of throughput calculations. The larger the size of thesubset is, the better the performance and the higher the computational load. However, inchoosing the subsets properly, the computational load can be significantly reduced whileproviding performance close to that of exhaustive search, as will be demonstrated vianumerical examples in Section 2.4.2. It is assumed that the train has higher or equalnumber of antennas than that of the serving BS. The proposed antenna selection basedtransmission scheme is summarized in Algorithm 2.

Algorithm 2 Two-phase antenna selection based adaptive transmission scheme forHST scenario.

Input:1: H = [H1, ...,HC] . MIMO channel estimate at the HST

Process:2: Select the best number of transmit antennas N∗t among a predefined set of options

N by choosing the one with the highest calculated throughput using (23), i.e.,N∗t = argmaxNt∈N TNt

3: Given the previously chosen number of transmit antennas N∗t , select the best antennapositions A among a predefined set of combinations A by choosing the one with thehighest calculated throughput using (23), i.e., A∗ = argmaxA∈A TNt (A)

4: Send the index of the chosen antenna positions A∗ to the BS.5: At the BS, use spatial multiplexing with the chosen antenna positions A∗ for data

transmission.6: At the train, estimate the MIMO channel H, and use the linear MMSE criterion for

data reception. Go to step 1.

The predefined set of options N depends on the maximum number of transmit antennasNtmax available and the level of complexity that can be accommodated. The basic ideabehind downsizing the number of antenna options is to choose the combinations that willmost probably provide the highest throughput performance. Given that the large size ofthe train can be utilized at the reception for the spatial separability of the transmittedstreams, having a high number of transmit antennas will most probably provide a highthroughput. In other words, there will be no point in checking the combinations withlow number of antennas, especially when the maximum antenna count is large. Forexample, if the maximum number of transmit antennas is 20, an antenna count of 10 or

48

less may be omitted. Thus, it appears to be beneficial to choose only high number ofantennas for the first phase of the algorithm.

In this first phase, only one antenna combination is selected for each antenna count.This selection could be random or the antennas could be chosen so that they are as farapart as possible from each other in order to provide as good a spatial separability aspossible. In the second phase of the algorithm different antenna combinations for thechosen antenna count in the first phase are compared, and the best one is chosen for thetransmission. If the antenna count chosen in the first phase is close to maximum (e.g.,19 out of 20), the performance difference between different combinations in the secondphase may be marginal since the corresponding radio channel characteristics are close toeach other. Hence, it is unnecessary to compare all these combinations to each other inthe second phase. The number of combinations can be reduced by randomly selectingonly a few of them to be compared. In conclusion, the basic idea behind Algorithm 2 toprovide high throughput performance with a low computational load is to choose highantenna counts for the first phase and select only a few antenna combinations of themfor the second phase. In general, most of the antenna counts and the correspondingcombinations are not relevant in terms of throughput performance. This is emphasizedwhen the maximum antenna count is high.

Intuitively speaking, the first phase may be more critical since the performancedifference between the best and the worst choices is probably higher than that in thesecond phase. Note that the feedback overhead with the implementation of Algorithm 2

is comparable to LTE dynamic transmission schemes, since the antenna combinationsare predefined and known at the BS and HST. Hence, only the index of the antennacombinations will be required for the feedback. Algorithm 2 can be designed so that thecomputational load is (mostly) lower than the LTE open-loop transmission scheme, aswill be demonstrated in Section 2.4.2.

Adaptive transmission scheme based on one-phase antenna selection

The second practical algorithm selects an appropriate number of transmit antennasand the proper positions of the active antennas simultaneously for spatial multiplexingtransmission. The algorithm operates as follows. The total throughput is first calculatedfor the maximum number of transmit antennas Ntmax using (23) or (24). Since spatialmultiplexing transmission is assumed, different transmit antennas can be ordered basedon their overall throughput performance.

49

Algorithm 3 One-phase antenna selection based adaptive transmission scheme forHST scenario.

Input:1: H = [H1, ...,HC] . MIMO channel estimate at the HST

Process:2: Calculate the total throughput for the maximum number of transmit antennas Ntmax

using (23)3: Set Nt ← Ntmax

4: Remove the antenna with the lowest throughput, and calculate the total throughputfor Nt −1 antennas using (23) to yield TNt−1

5: Compare the throughput values TNt and TNt−1

6: if (TNt > TNt−1) then7: Stop8: Choose Nt transmit antennas and the corresponding antenna positions9: else

10: Set Nt ← Nt −111: Go to step 512: end if13: Send the information of the chosen transmit antennas to the BS14: At the BS, use spatial multiplexing with the chosen transmit antennas for data

transmission15: At the train, estimate the MIMO channel H, and use the linear MMSE criterion for

data reception. Go to step 1

The transmit antenna with the lowest throughput is removed, and the total throughputis calculated again for Ntmax − 1 transmit antennas. The two calculated throughputvalues are compared. If the first value is higher, the maximum number of transmitantennas is chosen for spatial multiplexing based data transmission. Otherwise, thethroughput of Ntmax −1 antennas is compared to the throughput calculated for Ntmax −2antennas when the second worst antenna is removed. The process of reducing thenumber of transmit antennas one after the other based on the lowest per transmit antennathroughput is repeated until the total throughput is reduced. In practice, if the totalthroughput difference between the current and previous case is relatively small, it maybe worthwhile to continue the process until the performance difference is large enough.Therefore, there can be a predefined threshold α for the throughput difference, which

50

needs to be exceeded to stop the process, and select the previous number and positionsof the transmit antennas. The second antenna selection based transmission scheme issummarized in Algorithm 3. It will be shown in Section 2.4.2 that Algorithm 3 has aneven lower computational load than that of Algorithm 2. However, the feedback overheadis somewhat higher since the antenna positions are not chosen from a predefined set ofoptions.

2.4 Performance evaluation

This section provides a description of the simulators used in analyzing the throughputperformance of the HST communication scenario. The single-cell and multi-cellsimulation models are presented in Section 2.4.1 and the simulation results are presentedin Section 2.4.2.

2.4.1 Simulator description

In this section, the theoretical single-cell and multi-cell simulators are introduced, andthe main simulation parameters are presented.

Single-cell simulation model

The single-cell simulation model consists of a single BS with Nt transmit antennasserving a four-carriage HST. Each carriage is equipped with an MRN having Nr receiveantennas. The SNR for the MRN is defined as P/N0. Two different channel modelsare used to model NLOS scattering and LOS type of conditions with uncorrelated andcorrelated antennas, respectively. For the case of uncorrelated antennas, the Rayleighfading channel model is exploited. In this model, each channel coefficient is drawnfrom a Gaussian distribution with zero mean and unit variance. For the correlatedantennas, we have used the stochastic Weichselberger channel model [104]. In thiscase, the received signals for each path of the different antenna elements are to bespatially correlated. Hence, spatial correlation matrices are generated to reflect thespatial correlation among the antenna elements. The spatial correlation coefficients for

51

the BS and MRNs are generated according to

Rb =

1 ρb

1,2 · · · ρb1,Nt

ρb2,1 1 · · · ρb

2,Nt...

.... . .

...ρb

Nt ,1 ρbNt ,2 · · · 1

,Rr =

1 ρr

1,2 · · · ρr1,G

ρr2,1 1 · · · ρr

2,G...

.... . .

...ρr

G,1 ρrG,2 · · · 1

(25)

where G is defined as G = Nr and G = MNr for MRN non-cooperative mode and MRNcooperative mode respectively. The correlation coefficients are defined as

ρbi, j = Li jµ, ∀i = 1, ...,Nt ; l = 1, ...,Nt i 6= j

ρrk,l = Lkl µ, ∀k = 1, ...,G; l = 1, ...,G k 6= l

(26)

where µ denotes a correlation factor generated randomly from a uniform distribution inthe range (0,1), while Li j and Lkl are normalized path losses generated from the "RuralMacro" path loss model in [105]. The perpendicular distance between the BS and train,antenna spacing and mean angle of arrival (AoA) are taken into account in the calculationof the path losses. The mean AoA is derived from a uniform distribution in the range(−π/2,π/2). The overall channel was modeled according to the Weichselberger modelHc = UBS(ΩHiid)UT

MRN [104], which gives a better correlation structure than theKronecker channel model [106]. UBS and UMRN are the orthonormal eigenvectorsderived from the spatial correlation matrices for the BS and MRN, respectively. Thesymbol represents an element-wise multiplication and the coupling matrix Ω, whosestructure reflects the spatial arrangement of scattering objects, is derived from theeigenvalues of Rb and Rr. In each simulation, the performance is averaged over 1000independent channel realizations. The main simulation parameters for the case ofcorrelated antennas are listed in Table 1.

Multi-cell simulation model

This HST simulation model is based on a realistic wrap around multi-cell environmentwith a central layout of 57 cells (19 tri-sector antenna sites). This central layout iswrapped around with the copies of it to ensure uniform interference levels across the 57cells, as shown in Fig. 3. The simulation run consists of 1000 drops with 50 channelsamples per drop. 100 ground users are randomly distributed throughout the centrallayout at the start of each drop. The rail is laid across the central layout with a radius of5 km so that the closest distance to any BS along the rail is 50 m.

52

Table 1. Simulation parameters for theoretical scenario (with correlated antennas) [64]©2016, IEEE.

Parameters ValuesCarrier frequency 2 GHzNumber of channel realization 1000BS antenna spacing 0.5 wavelengthLength of each carriage 40 m⊥ distance (BS→ HST) 70 mMRN antenna spacing Length of carriage/number of MRN antennasNumber of BS/MRN antennas 4, 8, 12, 16correlation factor Uniform distribution (0,1)Angle of arrival Uniform distribution (−π/2,π/2)

The movement of the HST is modeled so that the HST moves incrementally along therail at the start of each drop. The ground users and MRNs on the HST are paired with57 cells providing the strongest received signal strength, which is calculated based ondistance dependent pathloss and angular antenna gain. The pairings remain constantthroughout each drop, therefore the handover processes are not considered as such.MIMO-OFDM system operating in FDD mode is considered with a channel bandwidthof 10 MHz according to the LTE standards [107].

At each transmission time interval (TTI), the resource scheduler at the BS distributesavailable resources between the active ground users and MRNs so that the overallthroughput is maximized and the resources are fairly distributed. The resources areshared based on proportional fair scheduling algorithm. First, the scheduler differentiatesground users from the MRNs and then splits the resources between ground usersand MRN backhaul links based on the number of active ground users and MRNs.A maximum of 50 percent of the resources are made available to the BS to trainbackhaul links due to the half duplex operation of the MRN. Scheduling relies onchannel information feedback, which may include CQI, PMI, RI and the proposedantenna selection indexing. Following the LTE framework, the channel feedbackinformation is obtained by the receiver/MRN and is made available to the scheduler witha periodicity of 6 ms interval and 2 ms delay. The values of the feedback periodicity anddelay depend on factors such as the granularity of the channel feedback informationsubband reporting, the capability of the user equipment, and higher layer message(e.g., radio resource control (RRC) Connection Reconfiguration, RRC ConnectionSetup) [107]. The calculated channel feedback information based on the channelcoefficients is made available at the receiver/MRN. Random errors are introduced

53

into the channel coefficients before calculating the channel feedback information, tomodel channel estimation errors as no specific channel estimator was implemented.The channel coefficients are generated based on the WINNER II channel model [95].This is a geometry based model that enables the separation of propagation parametersand antennas. Some of the parameters needed for the channel generation include thepropagation scenario, the speed and direction of the UE, number, height, and location ofthe BS and MS/MRN, antenna field patterns, system centre frequency, etc. Based on thedelayed channel feedback information, a modulation and coding scheme (MCS) is setfor each terminal at the BS for the next transmission.

A link to system (L2S) level interface is used to map the link level SINR values to thesystem level throughput performance. More precisely, the obtained SINRs are mappedto mutual information values using mutual information effective SINR metric (MIESM)link layer abstraction[108]. The MCS values determine the frame error probability (FEP)at the link to system level interface and define the transport block size, i.e., the numberof bits transmitted. The success of transmissions is identified by hybrid automaticrepeat request (HARQ) acknowledgements, which are determined in the system levelinterface and fed back to the BS after a delay. The number of bits transmitted forsuccessful transmissions and retransmissions are used in the throughput calculations.The described system level simulator is LTE compliant. The simulation parameters areset according to the guidelines established by the international telecommunicationsunion radiocommunication sector (ITU-R) for international mobile telecommunicationsadvanced (IMT-A) radio interface evaluation [109]. The main simulation parameters arelisted in Table 2.

2.4.2 Simulation results

In this section, the performance of different MIMO-OFDM transmission schemes isevaluated via numerical examples in single-cell and multi-cell simulation models. First,we compare the performance of the cooperative and non-cooperative MRN systems withthe increasing number of transmit and receive antennas in a simple single-cell scenario.Then, the performance of the LTE-based precoding schemes and the proposed simplifiedantenna selection algorithms are compared in a practical HST scenario. Finally, westudy how much the throughput performance can be improved by increasing the numberof transmit and receive antennas in the practical scenario.

54

5 10 15 20 250

10

20

30

40

50

60

70

80

90

100

110

120

SNR [dB]

Rate

[b/s

/Hz]

COP: Nt= 4; N

rTot= 4

COP: Nt= 8; N

rTot= 8

COP: Nt= 12; N

rTot= 12

COP: Nt= 16; N

rTot= 16

NonCOP: Nt= 4; N

rTot= 4

NonCOP: Nt= 8; N

rTot= 8

NonCOP: Nt= 12; N

rTot= 12

NonCOP: Nt= 16; N

rTot= 16

(a) i.i.d channel

5 10 15 20 250

10

20

30

40

50

60

70

80

90

100

SNR [dB]

Ra

te [b/s

/Hz]

COP: Nt= 4; N

rTot= 4

COP: Nt= 8; N

rTot= 8

COP: Nt= 12; N

rTot= 12

COP: Nt= 16; N

rTot= 16

NonCOP: Nt= 4; N

rTot= 4

NonCOP: Nt= 8; N

rTot= 8

NonCOP: Nt= 12; N

rTot= 12

NonCOP: Nt= 16; N

rTot= 16

(b) Correlated channel

Fig. 4. Achieved sum rate for cooperative and non-cooperative MRN system [64] ©2016,IEEE.

55

Table 2. Simulation parameters for practical scenario [64] ©2016, IEEE.

Parameters ValuesCarrier frequency 2 GHzSystem bandwidth 10 MHzTransmit power 46 dBmNumber of ground users 100Number of MRNs 8 (1 in each carriage)Propagation scenario suburban macro/rural moving networksInter-site distance (ISD) 1.3 kmTraffic model Full bufferMRN duplex mode half duplex FDDLength of each carriage 30 mTransmission scheme spatial multiplexingHARQ chase combiningReceiver type MMSEL2S interface metric MIESMTrain speed 300 km/h

In Fig. 4, the achieved sum rate is plotted against SNR for the cooperative (COP) andnon-cooperative (NonCOP) MRN systems with different numbers of transmit andreceive antennas in uncorrelated and correlated channels. The total number of receiveantennas at the train is denoted by NrTot = MNr. The simulation results show that thecooperation of MRNs is highly beneficial compared to non-cooperative scheme for allSINR values. The performance gain increases with the increasing number of transmitand receive antennas. The gains are even higher in correlated channel conditions, eventhough the absolute performance is somewhat reduced compared to the uncorrelatedchannels due to the reduction of spatial degrees of freedom. The superior performance ofthe cooperative scheme is because the overall MIMO channel from the BS to the entiretrain can be divided into parallel interference-free sub-channels with proper transmitand receive processing through the SVD. In the non-cooperative scheme, with theindependent data reception of each MRN, inter-MRN and inter-stream interferences limitthe throughput performance. In other words, the spatial separability of the transmittedsignals at the receiver side is more efficient for the cooperative scheme since it canexploit all the receive antennas and the physical length of the train. In conclusion, thetheoretical results imply that the MRNs in the train should cooperate if possible and theantenna arrays at the BS and the train should be as large as practically possible. Hencefor the rest of the result section, we only consider the cooperative scheme.

56

0 10 20 30 40 50 60 70 80 90 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mb/s]

C D

F

HST Speed: 300 km/h

HST Speed: 75 km/h

HST Speed: 3 km/h

Fig. 5. CDF plot of throughput for different HST speed [64] ©2016, IEEE.

Before examining the performance analysis for different transmission schemes, weshow the impact of different train speeds on the throughput of the cooperative scheme.Fig. 5 shows the cumulative distribution function (CDF) of the throughput for threedifferent train speeds with antenna configuration of Nt = 4 and Nr = 2 for each of theMRN. The results show the trend that throughput decreases as the speed increases. Thedifference in the throughput is mainly a result of the impact of the delayed feedbackof the channel information w.r.t. the channel and speed of the HST.1 Even though,due to the high speed, feedback information becomes outdated fast, the impact on thethroughput is small because the train is equipped with multiple MRNs, thereby forminga large antenna array. Due to the large antenna array, the received SINR is improved byan array gain and diversity gain. The array gain is due to multiple MRNs with multiplereceive antennas acting in a cooperative mode. The diversity gain which is due to thelarge antenna spacing is achieved by averaging over multiple independent paths. Hence,the probability that the overall gain is small is reduced and the impact of speed/delayedfeedback on the throughput is reduced.

1The effect of Doppler shift and handover at different vehicular speeds were examined in [110].

57

Fig. 6. CDF plot of throughput for fixed and adaptive transmission schemes with Nt = 4 andNr = 4 [64] ©2016, IEEE.

In all the following simulations, the total transmit power of a BS is assumed to be fixed(i.e., 46 dBm) and this power is equally divided among the transmit antennas, regardlessof the number of antennas used. Fig. 6 illustrates the CDF of the throughput for differenttransmission schemes at 300 km/h train speed. The following transmission schemes areconsidered

– Spatial multiplexing: fixed rank (S-MUX)– LTE closed-loop: fixed rank (LTE CL)– LTE open-loop: fixed rank (LTE OL)– LTE closed-loop: dynamic rank (LTE CL-dynamic rank)– LTE open-loop: dynamic rank (LTE OL-dynamic rank)– Spatial multiplexing: exhaustive search over all possible antenna combinations

(S-MUX exhaustive search).

The results show that all the fixed rank transmission schemes have nearly comparableperformance, regardless of the precoding. Thus, precoding-based schemes hardly

58

provide any gain over the S-MUX. However, the dynamic rank transmissions providesignificant performance gains compared to the fixed rank schemes. It can be seen thatthe LTE OL-dynamic rank scheme slightly outperforms the LTE CL-dynamic rankscheme. The superior throughput performance can be achieved by using the S-MUXwith exhaustive search. The results imply that using precoding may not provide anygains in a practical HST scenario. However, significant gains seem to be available byusing adaptive transmission schemes.

Fig. 7. CDF plot of throughput for adaptive transmission schemes with Nt = 4 and Nr = 4 [64]©2016, IEEE.

In Fig. 7, the CDF throughput performance of the proposed antenna selection (AS)algorithms are compared with the LTE OL-dynamic rank and the S-MUX exhaustivesearch transmission schemes. The proposed algorithms are represented as

– Alg.2 (2-phase AS)– Alg.3 (1-phase AS).

As can be seen, the proposed algorithms achieve an almost comparable performance tothat of the S-MUX with exhaustive search, yet they do this with a significantly reduced

59

computational load. The results also show that the proposed algorithms provide someperformance gains when compared to the LTE OL-dynamic rank transmission scheme.The computational loads involved for performing different adaptive transmissionschemes are evaluated in Table 32 in terms of per-stream throughput calculationsper-subcarrier.

Table 3. Number of per-stream throughput calculations [64] ©2016, IEEE.

Number of throughput calculationsNt 4 8 12 16 20LTE CL-dynamic rank 144 905 N/A N/A N/ALTE OL-dynamic rank 9 35∗ N/A N/A N/AS-MUX exhaustive search 28 1016 24564 524272 10485740Alg.2 (2-phase AS) 9.7 28.3 60.8 105.1 162.5Alg.3 (1-phase AS) 7.9 21 35.3 54 73

Note that in the case of the proposed algorithms, the average per-stream throughput needsto be calculated over the simulation runs since the number of throughput calculations isnot predefined, as in the case of all the other schemes. For all the transmission schemes,single layer transmissions were omitted.

One can see that the simple proposed algorithms require a significantly lowercomputational load than the S-MUX exhaustive search and CL-dynamic rank schemes. Itis worthwhile mentioning that one per-stream throughput calculation of an S-MUX-basedscheme requires a lower computational load compared to that of the precoding-basedscheme. Nevertheless, this is not taken into account in Table 3. The computational loadsof the proposed algorithms are slightly less than that of the OL-dynamic rank schemeexcept for Algorithm 2 with Nt = 4, in which the complexities are comparable. However,the difference increases as the number of antennas increases. The number of throughputcalculations is not presented for the LTE-based schemes when Nt > 8 since the LTEstandard only support Nt ≤ 8, and there are no codebooks defined for a higher numberof antennas. It is worth mentioning that the predefined antenna combinations in the firstand second phases of Algorithm 2 were chosen by following the principles described inSection 2.3.2 with the aim of keeping the computational load to a reasonable level. Inthe first phase, the number of antenna combinations |N | is related to the total number oftransmit antennas, and is given by |N |= (Ntmax/2)−1. The number of subsets in the

2Note that 8 transmit antennas are supported in the LTE specification for LTE CL but not for LTE OL.However, the computation for ∗LTE OL-dynamic rank as shown in Table 3 is based on a straightforwardextension from the 4 transmit antennas.

60

predefined set of antenna combinations A in the second phase ranges from 1 to 3 andthe choice of subset depends on the antenna count chosen in the first phase relative tothe total number of antennas. As an example, Table 4 presents the predefined antennacombinations for the case of Ntmax = 12. The numbers in the brackets refer to the indexesof the used antennas.

Table 4. Predefined antenna combinations for Algorithm 2 [64] ©2016, IEEE.

Antenna count 1st phase 2nd phase12 1,2,3,4,5,6,7,8,9,10,11,12 -11 1,2,3,4,5,6,7,8,9,10,12 1,3,4,5,6,7,8,9,10,11,1210 1,2,3,4,5,6,7,9,11,12 1,3,4,5,6,8,9,10,11,129 1,2,3,4,5,6,8,10,12 1,3,4,5,7,8,9,11,12

2,3,4,6,7,8,9,10,118 1,2,3,5,7,9,11,12 1,3,4,6,7,8,10,12

1,3,5,7,8,9,11,122,4,6,8,9,10,11,12

Inspired by the theoretical simulation results, we next examine how much the throughputperformance can be improved by increasing the number of transmit and receive antennasin a practical HST scenario. Fig. 8 shows the CDF of throughput performance for theS-MUX with fixed transmission scheme for various transmit and receive antenna arraysizes. It can be seen that the performance significantly improves with the increasingnumber of antennas. However, the performance gains are somewhat smaller when thenumber of antennas is high, especially for the low percentile region. This is mainlybecause increasing the number of antennas on the HST will reduce the antenna spacingbetween the antennas, thereby causing an increase in the spatial correlation of thechannel. Thus, the spatial separability of the transmitted streams becomes poorer.

The CDF throughput performance for the proposed antenna selection algorithmsusing the same number of transmit and receive antennas as in Fig. 8 is shown in Fig. 9.Comparing Fig. 8 to Fig. 9, the proposed algorithms show significant throughputperformance gains. The percentage increase in the throughput at different CDF pointscompared to fixed S-MUX transmission scheme is illustrated in Table 5.

Algorithm 2 and 3 have comparable performances in most cases. Algorithm 3 has alower computational load compared to Algorithm 2, however the feedback overheadis higher than in Algorithm 2 implementation. This is because the BS does not haveprior knowledge of which transmit antennas are active or inactive in Algorithm 3

implementation.

61

Fig. 8. CDF plot of throughput for fixed S-MUX transmission with increasing number ofantennas [64] ©2016, IEEE.Table 5. Performance gains of the proposed adaptive algorithms over fixed S-MUX scheme[64] ©2016, IEEE.

% increase in throughput performance compared to S-MUXCDF Alg.2 (2-phase AS) Alg.3 (1-phase AS)Nt 4 8 12 16 20 4 8 12 16 200.2 32% 45% 22% 42% 48% 35% 44% 30% 39% 42%0.5 6.3% 34% 19% 27% 22% 6.3% 34% 26% 19% 22%0.7 17% 20% 22% 22% 18% 17% 20% 25% 12% 17%0.9 30% 21% 9% 19% 11% 30% 17% 9% 18% 11%

S-MUX exhaustive search could not be simulated because of the large number ofthroughput calculations, which grows exponentially with an increase in the number ofantennas. The results imply that there are significant throughput gains available byincreasing the number of antennas and using the proposed low complexity algorithms.

62

50 100 150 200 250 300 350 4000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mbps]

C D

F

Nt= 8: Alg.2 (2−phase AS)

Nt= 8: Alg.3 (1−phase AS)

Nt= 12: Alg.2 (2−phase AS)

Nt= 12: Alg.3 (1−phase AS)

Nt= 16: Alg.2 (2−phase AS)

Nt= 16: Alg.3 (1−phase AS)

Nt= 20: Alg.2 (2−phase AS)

Nt= 20: Alg.3 (1−phase AS)

Fig. 9. CDF plot of throughput for the proposed adaptive transmission schemes with increas-ing number of antennas [64] ©2016, IEEE.

63

2.5 Summary

In this chapter, we considered a MIMO-OFDM HST communication system wherethe focus was on the backhaul link optimization with the aim of achieving throughputmaximization. This problem was approached from both theoretical and practicalperspectives. The theoretical results showed that cooperation between the MRNs andincreasing the number of antennas is highly beneficial. In a practical HST scenario, itwas shown that adaptive rank transmission schemes provide significant throughputgains compared to fixed rank schemes. However, adaptive schemes may require anextensive amount of computation, especially when the number of antennas is high.To significantly reduce the computational load, we proposed two low complexitytransmission schemes with simplified antenna selection and spatial multiplexing. Thepractical simulation results demonstrated that the proposed simple algorithms providealmost similar performance compared to the spatial multiplexing with exhaustive searchand improved the throughput performance compared to the LTE-based dynamic rankschemes. Furthermore, the computational load was significantly reduced. It was alsoshown that the throughput can be further improved by combining the proposed lowcomplexity algorithms with large antenna arrays.

64

3 Resource management in HSTcommunications

In this chapter, resource management is considered for moving relay node (MRN)equipped high speed trains (HSTs) in a regular cellular network. The focus is onmaintaining a guaranteed quality of service (QoS) for ground macro users (GMUs) whileensuring high data throughput on the HST utilizing an appropriate resource schedulingapproach. The impact of varying the number of MRNs on the throughput performance isalso evaluated. In Section 3.1, the system model is introduced. The scheduling problemwithin the LTE framework is presented in Section 3.2. Section 3.2 also describes the twohybrid scheduling algorithms and the different resource scheduling approaches. Theproposed algorithm for achieving the optimum number of MRNs to be used on an HSTis formulated in Section 3.3. The performance evaluation is presented in Section 3.4 anda summary is provided in Section 3.5.

3.1 System model

We consider a train with a two-hop network system architecture created with MRNsin a multi-cell system as shown previously in Chapter 2 (Fig. 3). The train consistsof Ncar carriages, each equipped with an MRN and an external antenna array that isevenly spaced along the length of the train and interior antennas to serve onboard users.The multi-cell system is modelled as a hexagonal layout consisting of 19 trisector cellswith replicas of the layout wrapped around the main layout as shown in Fig. 3. This isto ensure uniform interference levels across the main layout. Each sector operatingat 2 GHz band with a 10 MHz bandwidth has Nt transmit antennas transmitting on L

transmission layers and the MRN has Nr receive antennas. Signals from onboard usersinterfering with the MRN’s external antenna array are omitted. We also model GMUswith low mobility, each with Nrg receive antennas. WINNER II channel model [95] isadopted as the fast fading channel model.

In this scenario, MIMO-OFDM based system is considered, operating in frequencydivision duplexing (FDD) mode and assumes that the MRN operates in half duplexmode. For the transmission to the GMUs, transmit diversity transmission scheme isimplemented. Correspondingly, spatial multiplexing transmission scheme is used for the

65

MRNs since we can take advantage of the spatial dimension of the HST. Each GMU canonly communicate directly with the BS and users on the HST can only communicatewith the BS via the MRNs. The BS is equipped with a resource scheduler, whichallocates the available resources to the possible connections according to a schedulingmethod. Due to the half duplex operation of the MRNs, the MRNs are scheduled atevery second time slot. The receive signal vector for the MIMO transmission yc ∈ CN

for subcarrier c can be expressed as

yc = Hcsc +K

∑k=1

Hk,csk,c +nc (27)

where Hc ∈ CN×Nt is the channel matrix for subcarrier c of the desired signal withdimension N ∈ Nr,Nrg and sc ∈CNt is the desired transmit signal vector per subcarrier.Hk,c and sk,c are the channel matrix and the transmit signal vector from interfering sectork to the MRN on subcarrier c, K is the total number of the interfering BSs, and nc isthe additive white Gaussian noise vector. The transmitted signal vector is estimated assc = Wcyc, where Wc = (HcHH

c +Rc)−1Hc is the optimal linear minimum mean square

error (MMSE) filter and Rc = ∑Kk=1 Hk,cΣNt H

Hk,c +N0IN is the interference plus noise

covariance matrix. ΣNt is a Nt-dimensional diagonal matrix whose diagonal elementsequally share the total transmission power pc,k from the kth interfering sector. Thecorresponding MRN’s SINR per transmission stream for subcarrier c at the output of theMMSE receiver can be expressed as

Γc,l =pc,l |wH

c,lHc|2L

∑j=1, j 6=l

pc, j|wHc, jHc|2 +wH

c,lRcwc,l

(28)

where Hc is the estimate of Hc, pc,l denotes the transmission power for the lth stream atsubcarrier c and wc,l denotes the lth column of Wc.

3.2 Resource scheduling approach

This section describes the resource scheduling between MRNs on the HST and theGMUs. The resource scheduling problem is presented in Section 3.2.1 within the contextof the LTE framework. Two proposed hybrid scheduling algorithms and schedulingapproaches are developed in Section 3.2.2.

66

3.2.1 Resource scheduling problem

GMUs and MRNs are associated with any of the 57 BSs in the central layout based onvariation in the instantaneous channel conditions, shadow fading and distance-dependentpath loss. Hence, with the possibility of having more than one GMU/MRN associatedwith a BS, scheduling is achieved based on the channel quality of each GMU/MRN.Scheduling ensures that the available radio resources are shared with as many usersas possible, while still satisfying GMU/MRN QoS requirements. The throughput ofeach GMU/MRN and the throughput of the entire network are affected by the typeof scheduler used. The type of scheduling method used also influences the impact ofthe HST on the throughput of the GMUs. Hence, evaluating the efficiency of differentscheduling methods as it relates to the GMUs and MRNs, is required.

We consider an LTE-A system with a radio frame structure, so that each radio frameis divided into 10 subframes. Resources are allocated to users on a subframe basis atevery 1 ms transmission time interval (TTI), (i.e., the scheduler allocates resourcesto GMUs/MRNs in minimum portions of two consecutive resource blocks (RBs)),due to signaling overhead limitations. One RB consists of 12 adjacent subcarriersand corresponds to 0.5 ms (i.e., 1

2 of a subframe) in the time domain. In any givencell, GMUs are always allocated resources at each TTI. However, MRNs are allocatedresources at every second TTI, since we assume half duplex functionality of the MRNs.Let us assume a time slot t equals 1 TTI (1 ms) and the overall system bandwidth isB MHz. With a scheduling block (SB) corresponding to two consecutive RBs, thebandwidth of each SB is B

NSB, where NSB = 2NRB is the number of SBs within a specified

bandwidth and NRB corresponds to the number of RBs. In each TTI, scheduling isdone across NSB scheduling blocks and we assume a uniform distribution of the totaltransmit power across all subcarriers. The capacity of the rth scheduling block for themth GMU/MRN on the t th TTI that establishes a connection link is given by

Cm,r(t) =B

NSBlog2

(1+

Cr

∑c=1

L

∑l=1

Γc,l

)(29)

where Cr is the number of subcarriers in the rth scheduling block. Let 1t,m denote anindicator function that describes GMUs/MRNs that are eligible to be allocated resourcessuch that

1t,m =

1 if (tind ∈ [1,2]∧ m ∈ GMU)∨ (tind ∈ [1]∧ m ∈MRN)

0 Otherwise

67

where tind = (t mod 2)+1 is an index that ensures an MRN is not scheduled on everysecond TTI and m is the GMU/MRN allocated resources on the rth scheduling block atthe t th TTI. The total achievable rate up to T TTIs in the bth cell can be calculated as

Rtot(b) = ∑m∈U

T

∑t=1

NSB

∑r=1

1t,mCm,r(t) (30)

where U is the set of GMUs/MRNs associated with the same BS. The overall throughputacross the network, i.e., ∑b∈B Rtot(b) is maximized using a scheduling method, where Bis the set of BSs within the network. However, Rtot(b) is a combination of the through-puts for GMUs and MRNs, i.e., Rtot(b) = Rgmu(b)+Rmrn(b). Therefore, combinedscheduling of both GMUs and MRNs may not necessarily ensure fairness between thetwo groups of users. To ensure adequate fairness between the two groups of users withminimal impact on the throughput, we analyze the performance of joint and separatescheduling of GMUs and MRNs using two hybrid scheduling methods.

3.2.2 Resource allocation methodology

Hybrid scheduling algorithms

Here, we consider two hybrid scheduling algorithms based on existing algorithms withthe aim of optimizing the scheduling performances.1. Round robin with max rate (RRMR): The RRMR scheduling algorithm is a combi-nation of round robin scheduling algorithm and maximum rate scheduling algorithmto bring about a balance between fairness and performance. The RRMR schedulingalgorithm presented in Algorithm 4 brings a balance between fairness between MTs andscheduling MTs with the highest achievable rate. For every TTI, MTs are scheduledacross NSB scheduling blocks. The MT with the best channel condition is first allocatedresources on a corresponding fixed set of scheduling blocks nmax

SB , then the MT withsecond best channel condition is allocated resources on the next available fixed set ofscheduling blocks nmax

SB . This process continues for all MTs associated with the BS. Afterall the MTs have been allocated resources, the first MT with the best channel conditionis assigned nmax

SB more scheduling blocks. This process continues until there are no moreavailable scheduling blocks or until the target rate of each MT is reached. The principleis to allocate a small number of SBs to all users cyclically in a predetermined orderbased on the maximum rate. Hence, the total SBs are shared as evenly as possible on abest performance basis, without unduly reducing the throughput.

68

Algorithm 4 RRMR scheduling algorithm.Initialization:

1: t← 0 . Initializing the scheduler based on TTIProcess:

2: t← t +1 . The t th TTI3: tind ← (t mod )2+1 . Assigning an index number to each TTI4: U 3 MTs available to share resources on the t th TTI5: Rm,r(t) = 1t,mCm,r(t) . Achievable rate on all SBs for each MT6: (m∗,r∗) = argmax

m,rRm,r(t)

7: Exclude the r∗ SB . Since r∗th SB is allocated to best MT m∗.8: n← 0 . nmax

SB :fixed set of SB to be scheduled to a MT.9: for n≤ nmax

SB do10: n← n+111: if (Allocated resources for m∗ < required rate for m∗) then12: r∗∗ = argmax

rRm∗,r(t) . The next best SB r∗∗

13: continue14: else15: m∗ 6∈U . Exclude m∗ from the set.16: if U 6= then17: if free SBs still available then18: break and go to step 619: else20: break and prepare for next TTI. go to step 221: end if22: else23: if (Allocated resources for each MT ≥ Required rate for each MT) then24: break and prepare for next TTI. go to step 225: else26: U 3 MTs available to share resources on the t th27: return to step 1728: end if29: end if30: end if31: end for

69

2. Modified proportional fair (MPF): The MPF scheduling algorithm is a slightmodification of the proportional fair (PF) algorithm. The proportional fair schedulingalgorithm provides a balance between optimized throughput and fairness between theMTs. The balance provided by the algorithm is as a result of taking into account thecurrent channel conditions and transmission history of the MTs. In addition to thePF, the MPF scheduling presented in Algorithm 5 also provides a balance betweenfairness and performance so that the selected scheduling block for an MT is based onthe MT with the best available rate on the scheduling block and the minimum averageavailable rate of all the MTs associated with the BS. Therefore, at each TTI, resourcesare allocated to MTs with the relatively best channel condition so that an MT is allocatedresources on a scheduling block according to

m∗ = argmaxm

Rm,r∗(t)Rm,r∗(t)+ ε

(31)

where Rm,r∗(t) = 1t,mCm,r∗(t) is the fed back data rate for MT m, with the selectedscheduling block r∗ described as

r∗ = argminr

(∑

m∈URm,r(t)

)/|U|. (32)

Rm,r∗(t) is the moving average throughput on the rth scheduling block for user m and ε

is a small positive constant to prevent the error of dividing by zero. The moving averagethroughput (transmission history) is calculated over a certain proportional windowlength Pw and updated as

Rm,r∗(t +1) =(

1− 1Pw

)Rm,r∗(t)+

1Pw

Rm,r∗(t). (33)

Scheduling approaches

We consider the joint and disjoint scheduling approaches w.r.t. the two groups of MTs,i.e., GMUs and MRNs.1. Joint Scheduling: In the joint scheduling approach, the GMUs and MRNs arescheduled together. All users associated with a BS are given equal priority. Hence, forboth scheduling algorithms, the set U which consist of MTs available to share resourcesinclude both GMUs and MRNs.

70

Algorithm 5 MPF scheduling algorithm.Initialization:

1: t← 0 . Initializing the scheduler based on TTIProcess:

2: t← t +1 . The t th TTI3: tind ← (t mod )2+1 . Assigning an index number to each TTI4: U 3 MTs available to share resources on the t th TTI5: Rm,r(t) = 1t,mCm,r(t) . Achievable rate on all SBs for each MT6: Obtain r∗ from (32)7: Obtain m∗ from (31)8: Exclude the r∗ SB . Since r∗th SB is already allocated.9: if free SBs still available then

10: if (Allocated resources for each MT ≥ Required rate for each MT) then11: prepare for next TTI. go to step 212: else13: return to step 614: end if15: else16: prepare for next TTI. go to step 217: end if

2. Disjoint Scheduling: The disjoint scheduling approach identifies and separates thetypes of MTs before applying the scheduling algorithm. Hence, the GMUs are scheduledseparately from the MRNs. However, scheduling priority has to be established. The ideaof scheduling with priority is to either schedule the GMUs first or the MRNs first. Theseparate scheduling of the GMUs and MRNs provides the ability to utilize differentscheduling methods for each group of users, thereby enhancing the flexibility of theachievable rate and fairness between the groups.

3.3 Optimum cooperative MRN deployment

A very promising solution in the HST scenario that ensures high throughput performanceis to equip each carriage with an MRN [27, 65]. A contributing factor to improvedperformance is the cooperation of the MRNs, which overcomes the limitations ofthe half duplexing operation of individual MRN. This option is favorable to mobile

71

operators, since the relay system is under the full control and management of the cellularnetwork. The feasibility and performance gains of the use of an MRN in an HSTover direct transmission to the OBUs was examined in [111], where issues of groupmobility, local service support, multi-RAT and RAN sharing were addressed. Resultsfrom [65] have shown that transmission through the use of multiple MRNs to OBUsachieves much higher throughput performance compared to direct connection or singleMRN link. Increasing the number of receive antennas on each of the MRNs furtherimproves the OBUs’ throughput, which directly implies that the OBUs’ throughputis dependent on the backhaul link throughput, i.e., BS-to-train link. Backhaul linkthroughput enhancement has been examined in [66] using transmit antenna positioning.

However, the use of multiple MRNs, where each carriage of the HST is equippedwith an MRN, can contribute significantly to the capital expenditure of the network. Ingeneral, most mobile network operators seek ways of ensuring high data throughputwhile reducing energy consumption and the cost of mobile cellular network infrastructure.Hence, in order to address the challenge of high capital expenditure while ensuring highdata throughput, an iterative based algorithm is proposed. The algorithm is developedbased on the assumption that the cost of an MRN is much higher than the cost ofadditional antenna elements. Hence, a trade-off between reducing the number ofMRNs on an HST and increasing the number of receive antennas is established. Theperformance gains from multiple backhaul links and diversity gains from the number ofreceive antennas are exploited to achieve the optimum number of MRNs (#MRN∗) andreceive antennas (N∗r ) to be used on an HST.

The iterative algorithm summarized in Algorithm 6 begins with a reference scenariowhere the HST is equipped with an MRN in each of the carriages and the number ofreceive antennas is set to two so as to ensure MIMO transmission. The number ofreceive antennas on the MRN is increased until a predefined target data rate can beachieved. It should be noted that the target performance metric can either be a predefinedcondition number or predefined level of channel correlation. Then the total number ofMRNs is reduced by one and the difference between the previous throughput (T re f

Nr)

and the current throughput (T newNr

) determines whether the number of receive antennasshould be increased or a further reduction in the number of MRNs is required. Thesymbol δ is a threshold that defines the level of precision chosen. The symbol β definesthe minimum number of MRNs required on the HST, while α prevents an excessiveincrease in the number of receive antennas. A flow chart implementation with emphasison the center of the cell, (i.e., 50%-tile CDF throughput) is given in Appendix 1.

72

Algorithm 6 Algorithm to optimally select number of MRNs and receive antennas.1: Input: Obtain number of carriages Ncar and minimum target rate Ttar.2: Initialization: Set number of MRNs M← Ncar and number of receive antennas

Nr← 2. Obtain MIMO channel estimate across the network layout and calculatethroughput TNr .Process:

3: while (TNr < Ttar) do4: Nr← Nr +1 . Increase Nr until the minimum target rate is achieved5: Calculate TNr

6: end while7: T re f

Nr← TNr . T re f

Nr: Reference throughput.

8: antAdd← 0. . Setting counter for Nr added.9: M←M−1. . reducing the number of MRN.

10: Calculate T newNr

. T newNr

: New throughput with updated M, Nr.11: if (T new

Nr−T re f

Nr)≤ δ then

12: if M ≤ β then13: go to Output14: else15: T re f

Nr← T new

Nr; then go to step 9

16: end if17: else18: Nr← Nr +1 . Increasing the number of receive antennas Nr

19: antAdd← antAdd +1 . Number of receive antennas added20: if antAdd > α then21: M←M+1 . Return M to previous value.22: Nr← Nr−1 . Return Nr to previous value.23: go to Output24: else25: T re f

Nr← T new

Nr; then go to step 10

26: end if27: end if

Output:28: Set #MRN∗←M. . MRN∗:optimum number of MRNs on HST.29: Set N∗r ← Nr. . N∗r :optimum number of receive antennas on HST.

73

Note that the median throughput baseline can be changed to reflect the region ofinterest, for example the cell-edge. With the reduction of the number of MRNs (i.e.,#MRN∗ < Ncar), we propose the implementation of a distributed antenna system (DAS)on the access link of the HST so that at least one remote antenna unit (RAU) is presentin each carriage, and the MRNs are able to cooperate in a coordinated fashion.

3.4 Performance evaluation

This section provides a description of the simulator used in analyzing the performanceof the MRNs and the GMUs as described in Section 3.4.1. The simulation results arepresented in Section 3.4.2.

3.4.1 Simulator description

An LTE-A based system level simulator is used in the performance evaluation in thischapter. The LTE-A based system level simulator is configured to follow the guidelinesestablished by the international telecommunications union radiocommunications sector(ITU-R) for IMT-A radio interface evaluation [109]. The simulation parameters are setto closely follow the LTE system. The central layout with an inter-site distance of 1.3km is modelled similar to that used in Chapter 2 (Fig. 3) and the antenna configurationsfor Nt , Nr, and Nrg are set to 4, 4 and 2, respectively. The simulation runs consist of1000 drops, with 100 GMUs randomly distributed across the central layout at the start ofeach drop. A track with a radius of approximately 5 km is placed across the centrallayout so that the minimum distance between the track and any BS is 50 m. At thestart of the first drop, the train is positioned on one end of the track and as the drops goon, the train moves along the track to the other end. At the start of each drop, each ofthe GMUs and the train equipped with MRNs are paired with 57 cells that provide thestrongest received signal strength. The received signal strength is calculated based ondistance dependent path-loss and angular antenna gain.

Scheduling the mobile users (GMUs and MRNs) at the BS relies on channel stateinformation (CSI) feedback, i.e., the channel quality indicator (CQI), which is derivedfrom the mobile users and is made available at the BS after a delay. A target rateof 10 Mbps is set for the GMUs, and a full buffer traffic model is considered for theMRN with the assumption that the number of users in the train may be large. A link tosystem (L2S) interface is employed so that the SINRs obtained are mapped to mutual

74

information values using mutual information effective SINR metric (MIESM) linklayer abstraction [108]. A modulation and coding scheme (MCS) is set for each mobileuser and the MCS values determine the frame error probability (FEP) at the link tosystem level interface and the transport block size, i.e., the number of bits transmittedfor throughput calculations. Successful transmissions/retransmissions are identified byhybrid automatic repeat request (HARQ) acknowledgements, which are determinedin the system level interface and fed back to the BS after a delay. For mobile userswith successful transmissions/retransmissions, the number of correctly received bits iscalculated and used in throughput calculations.

3.4.2 Simulation results

The following presents the simulation results according to our model for downlinktransmission. Fig. 10 gives the cumulative distribution function (CDF) of the GMU/MRNthroughput for different scheduling approaches using the RRMR scheduling algorithm.The following scheduling approaches are considered:

– RRMR algorithm giving the MRNs priority (RRMR MRN)– RRMR algorithm giving the GMUs priority (RRMR GMU)– RRMR algorithm with joint scheduling (RRMR Joint).

With the RRMR GMU approach, the total GMU throughput in Fig. 10(a) improvessignificantly compared to the joint and MRN priority approaches. However, in Fig.10(b), the MRNs suffer an outage for a significant part of the journey. On the otherhand, when the GMUs and MRNs are jointly scheduled (i.e., RRMR Joint), the totalMRN throughput in Fig. 10(b) is significantly improved with no outage. But the totalGMU throughput for RRMR Joint approach suffers a significant reduction, even whencompared to RRMR MRN approach. This is because, in a given cell the MRNs are morelikely to have better channel conditions than the GMUs and as a result, the GMUs aremost often not allocated the best resources. The MRNs will most often have betterchannel conditions because the antenna heights are much higher than the GMUs’ sincethe antennas are mounted on the top of the train. In a case where the GMUs are muchcloser to the BS than the MRNs, the GMUs may have better channel conditions. Thisexplains what happens, when RRMR MRN and RRMR Joint approaches are comparedat the cell edge in Fig. 10(b).

75

0 1 2 3 4 5 6 7 8 9 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mb/s]

CDF

(a) GMU Throughput

RRMR MRNRRMR GMURRMR Joint

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mb/s]

CDF

(b) MRN Throughput

RRMR MRNRRMR GMURRMR Joint

Fig. 10. CDF throughputs with RRMR scheduling method [82].

The MPF scheduling method is applied using the scheduling approaches discussedwith the CDF throughputs shown in Fig. 11. The following scheduling approaches areconsidered:

– MPF algorithm giving MRNs priority (MPF MRN)– MPF algorithm giving GMUs priority (MPF GMU)– MPF algorithm with joint scheduling (MPF Joint).

As expected, with the MPF GMU approach, the total GMU throughput (Fig. 11(a))is significantly improved compared to the other two approaches, but the total MRNthroughput (Fig. 11(b)) is poor with outages. The total MRN throughput for the othertwo approaches shows that giving MRNs priority (i.e., MPF MRN approach) canachieve about 5 Mbps improvement compared to joint scheduling (MPF Joint). However,considering Fig. 11(a), the total GMU throughput for the MPF Joint approach achievesa large improvement in compared to the MPF MRN approach from the 50% CDF point.

76

0 1 2 3 4 5 6 7 8 9 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mb/s]

CDF

(a) GMU Throughput (MPF)

MPF MRNMPF GMUMPF Joint

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput [Mb/s](b) MRN Throughput (MPF)

MPF MRNMPF GMUMPF Joint

Fig. 11. CDF throughputs with MPF scheduling method [82].

But at the cell edge, the MPF MRN approach performed better than the MPF Jointapproach. This is because in the MPF Joint approach, the MRN which is jointlyscheduled brings an increase in interference to the cell edge GMUs. From Fig. 10and 11, it can be seen that scheduling MRNs first, brings a fair balance to the GMUand MRN throughputs. Due to the ability to separate the scheduling of each groupof users, we can use different scheduling algorithms for each group of users with theaim to improve performance on both throughputs. Fig. 12 shows the throughput CDFperformances of GMUs and MRNs, where an MPF scheduling algorithm is applied forthe GMUs and maximum rate (MR) scheduling algorithm is applied for the MRNs. Wecall this the MPF-MR scheduling algorithm and compare scheduling the GMUs first(MPF-MR GMU) to scheduling the MRNs first (MPF-MR MRN). The results show thatto avoid outages on the train, it is better to schedule the MRNs first.

The total MRN throughput as seen in Fig. 12(b) shows that giving MRN prioritywith the MR algorithm can further improve the throughput performance by about 5 Mbpscompared to the MPF MRN scheduling approach in Fig. 11(b) with a minimal impact

77

on the total GMU throughput performance (Fig. 12(a)) when compared with MPFJoint scheduling approach in Fig. 11(a). The gain made by using the MR schedulingalgorithm on the MRNs is because multiple carriages equipped with an MRN in eachcarriage can be made to cooperate in a coordinated fashion.

Fig. 12. CDF throughput with MPF scheduling for GMU and MR scheduling for MRN [82].

In Fig. 13, we examine the impact of reducing the number of MRNs in an HST whileincreasing the number of receive antenna elements for each MRN so that the totalnumber of MRNs and receive antennas are equal for the given simulation, i.e., the totalnumber of receive antennas on the HST are the same for the three CDF curves. Theantenna configuration (antconf) is represented as Nt ×Nr. The throughput CDF plotsshow that more MRNs with reduced receive antenna elements can achieve a higherthroughput performance than an HST with a lower number of MRNs with increasedreceive antenna elements. The spatial diversity gained from the use of more MRNs isachieved from the HST connecting to multiple BSs, and efficient bandwidth utilizationis achieved with the coordination and cooperation of more MRNs. On the other hand,the spatial diversity gained from having more receive antenna elements does not provideas large a gain as the spatial diversity gained from connecting to multiple BSs.

78

0 10 20 30 40 50 60 70 80 90 1000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Throughput (Mb/s)

CD

F

#MRN: 2; antconf: 4 X 16#MRN: 4; antconf: 4 X 8#MRN: 8; antconf: 4 X 4

Fig. 13. CDF plot of MRN throughput with equal total number of MRN and receive antennas[93] ©2016, IEEE.

Fig. 14. CDF plot of MRN throughput with varying number of MRN and receive antennas [93]©2016, IEEE.

79

Table 6. Set values in the proposed algorithm.

Reference values(Initial) Threshold

Nr Number of MRNs ISD (km) δ α β

4 8 1.7 0 3 2

The throughput CDF of the MRN system for varying numbers of receive antennasand MRNs are shown in Fig. 14. The figure shows that when the number of receiveantennas is significantly increased with a significant reduction in the number of MRNs,the throughput CDF performance is worse compared to the case with an MRN in eachof the carriages of the HST with a reduced number of receive antennas.

Fig. 15. CDF plot of MRN throughput for the first and second iteration of the proposedalgorithm [93] ©2016, IEEE.

However, a proportionate increase in the number of receive antennas and a reduction inthe number of MRNs show that an improved throughput performance can be achieved.

The optimal tradeoff between the number of MRNs and the number of antennas isprocessed in the throughput CDF of the MRN system using the proposed algorithm inFig. 15 and 16 with the reference scenario and threshold set as shown in Table 6.

80

Fig. 16. CDF plot of MRN throughput for the third and fourth iteration of the proposed algo-rithm [93] ©2016, IEEE.

The left plot of Fig. 15 shows the throughput for first iteration of the proposed algorithm.Reducing the number of MRNs by one while maintaining the same number of receiveantennas shows a 10.5% reduction in throughput at the 50% CDF point. To complete thefirst iteration, the number of receive antennas is increased by one and the results show a3% increase from the initial throughput at the 50% CDF point. Hence, the process iscontinued for the second iteration with results shown in the right plot of Fig. 15. Theresults show that two additional receive antennas were required for the algorithm tocontinue the process. The iterative process of the algorithm terminated at the fourthiteration shown in the right plot of Fig. 16, since the additional receive antennas reachedthe threshold limit (α). Therefore, 5 MRNs and 9 receive antennas will be the proposedvalues to be used by the algorithm for an HST with 8 carriages. It should be noted thatthe outcome of the algorithm also depends on the ISD. This algorithm ensures a lowernumber of MRNs without any performance loss in terms of the throughput.

81

3.5 Summary

In this chapter, two hybrid resource scheduling algorithms were examined in the contextof two different groups of MTs, the GMU and the MRN mounted on the HST. In orderto address the challenge of efficiently and fairly sharing resources between the GMUsand MRNs, RRMR and MPF scheduling algorithms were analyzed based on joint anddisjoint scheduling approaches. In the joint scheduling approach, optimal performanceand fairness can not be achieved since the priority for scheduling is only based onsignal quality. However, the scheduling priority in the disjoint scheduling approach canbe based on the combination of signal quality and the type of MT. Furthermore, thedisjoint scheduling approach gives the flexibility to use a mix of different schedulingalgorithms to improve the fairness and throughput performance. Additionally, the impactof reducing the number of MRNs mounted on an HST was examined with the proposalof an iterative algorithm to efficiently select the number of MRNs and receive antennasthat should be mounted on an HST, while still maintaining a high throughput. The effectof reducing the number of MRNs compared to reducing the number of receive antennason the throughput performance was also examined. The reduction in the number ofMRNs can lead to significant cost savings.

82

4 Beamforming at higher frequency bands forHST communications

In this chapter, higher frequency bands (HFBs) are considered for high speed train(HST) communication networks. The range of frequency bands under consideration isbetween 6 and 100 GHz. The focus is on designing transmit and receive beamformers atHFBs for HST communication networks. However, before the beamforming design, weexamine the feasibility of the use of HFBs for high speed railway communications. InSection 4.1, the background and suitability studies on the use of HFBs for HST networksis examined. The system model for the HST network is introduced in Section 4.2. Theproposed beamforming scheme as applied to the HST environment is formulated inSection 4.3 with the performance evaluation presented in Section 4.4. The chapter issummarized in Section 4.5.

4.1 Suitability of higher operating frequencies for railwaycommunications

In this section, a feasibility study on the use of HFBs for broadband communicationsin HSTs is presented, in which the railway radio environment and the impact of highvelocities are taken into account in the modification of the modulation scheme beingused. A brief background to the characteristics of the railway propagating environmentis provided in Section 4.1.1 and a modified HFB frame structure is presented in Section4.1.2.

4.1.1 The railway deployment scenario

The railway deployment scenario aims to achieve continuous coverage and high data ratesfor MTs inside an HST along the rail track. In this context, the MTs are a combinationof essential train communication devices which require high reliability and onboardpassengers who require high data rates [112]. To guarantee reliable communications inthe HST network, knowledge about the high speed railway environment has been shownto be instrumental [113]. The propagation characteristics of the railway environmenthave been observed in a number of studies in channel measurements and analyses

83

at microwave frequency bands [114–120]. For most part of the railway propagatingenvironment, a strong LOS component is present and there are few multipaths becauseof small scatterers and reflections, which is closely similar to the Rural macro-cell(RMa) propagation scenario described in [95].

BS

BS

BS

RRH RRH RRH RRH

RRH RRH

1732 m577 m 577 m

Fig. 17. HST scenario with RRHs to ensure LOS propagation [94] ©2017.

Considering HFBs for the HST scenario and maintaining the network layout formicrowave bands so as not to increase the strain on the existing challenge of frequenthandovers, the LOS component will be lost especially at the cell-edges and NLOScomponents will be negligible due to fewer scatterers. However, the use of remote radioheads (RRHs) [121] and moving relay nodes (MRNs) on the HST can be used to ensurethe presence of LOS propagation as shown in Fig. 17.

Previous studies in [64] identified the critical differences between the railwaycommunication system and the conventional cellular system with proposed transmissionschemes for HST scenarios using a two-hop network architecture. A single hop networkarchitecture was adopted in [122], where advanced collaboration schemes were used toimprove the throughput. However, with increasing data rate requirements, the use ofadvanced transmission schemes and techniques at microwave bands are not enough tomeet the data rate demand. Hence, the large available bandwidth at higher operating

84

frequency bands is seen as a potential way to meet the data rate demand in the railwaydeployment scenario.

Hence, the impact of high mobility and additional pathloss must be taken intoaccount in the choice of HFB to ensure proper communication network planning. Also,the increased impact of the delay spread and Doppler shift/spread must be taken intoaccount in the development of the OFDM frame structure when HFBs are considered.

4.1.2 HFB frame structure for HST and effect of large bandwidth

The design of a 5G new radio physical layer is heavily influenced by the requirementsfor high data rates, improved spectral efficiency, and the availability of larger channelbandwidths. To fulfil these requirements, OFDM frame structure is proposed as a goodcandidate. Also, some advanced technical features of LTE-A are being considered for5G systems, where 5G design exploits ways to combine existing 4G LTE networks withcapabilities provided by 5G [123]. Hence, the existing LTE framework is used as abaseline in the development of the HFB frame structure for HST. The development ofthe frame structure for HST is vital in predicting accurate lower bounds on SNR/SINRfor a given target rate. We propose the use of the modified alpha Shannon formula[124] to facilitate accurate benchmarking of the OFDM structure so that the physicallayer parameters are taken into account in deriving the minimum SNR/SINR value forsuccessful transmission for a given target rate. The minimum SNR/SINR for successfultransmission can be expressed as

Γmin = β

(2

RtarαB −1

)(34)

where Rtar is the target rate and B is the transmission bandwidth. The symbol β

represents the SNR efficiency factor, which is partly a function of the used modulationand coding scheme and performance aspect of the receiver algorithms. Hence, the SNRefficiency factor can be extracted from curve fitting to link-level simulations [124]. Thesymbol α represents the system bandwidth efficiency factor. The bandwidth efficiencyfactor quantifies the available bandwidth that can be used for transmission. This takesinto account the overhead of the following physical layer parameters [125]; the cyclicprefix, the pilot assisted channel estimation, common control channels and adjacentchannel leakage ratio (ACLR) requirements. Based on these parameters, the bandwidthefficiency factor is expressed as

α = (1−Ncp)(1−Npilot)(1−NL1/L2)(1−NACLR) (35)

85

where Ncp, Npilot , NL1/L2 and NACLR represent the percentage overhead of the corre-sponding physical layer parameters.

Due to the high speed of the train and the shorter wavelengths of the higher operatingfrequency bands, the OFDM frame structure needs to be defined for HFBs so that thesubcarrier spacing ∆ f is adjusted with the following constraints

∆ f 1/TCP and ∆ f fD

where TCP and fD represent the length of cyclic prefix and the maximum Doppler shift,respectively.

Without a loss of generality, the 5G new radio frame should maintain the LTE radioframe of 10 ms with a subframe of 1 ms and a time slot of 0.5 ms. The sub-carrierspacing for the HFBs is also constrained so that the sampling rate is a multiple orsub-multiple of the WCDMA chip rate of 3.84 Mcps. Furthermore, the number ofOFDM symbols in one time slot will be affected by changes in sub-carrier spacing sincethe length of the useful symbol is given as Tu = 1/∆ f . The number of OFDM symbolsis expressed as

Nsym =Tslot − (Tu +TCP1)

Tu +TCP2

+1 (36)

where Tslot is the time slot, TCP1 is the length of cyclic prefix for the first symbol in eachtime slot and TCP2 is the length of cyclic prefix for the rest of the symbols. Hence, Ncp isaffected by the value of Nsym and given as

Ncp =TCP1 +(Nsym−1)TCP2

Tslot. (37)

Still using the LTE framework as a baseline, transmission is scheduled in resourceblocks (RB), each of which consists of 12 consecutive sub-carriers for a duration ofone time slot. The pilot symbols for channel estimation are inserted in the OFDMtime-frequency grid with a time domain spacing of four symbols and a frequencydomain spacing of six sub-carriers. Therefore, Npilot is derived as the ratio between thenumber of pilot symbols and the number of resource elements in a time slot. We assumethat the values of NL1/L2 and NACLR are fixed irrespective of the carrier frequency used,since they are independent of the carrier frequency. Table 7 shows the derived 5G newradio physical layer parameters for different operating frequency bands using the LTEframework [126] as a baseline and assuming a maximum mobile speed of 500 km/h.

As seen in Table 7, we assume the same cyclic prefix length across all frequencybands despite the changes in the symbol length Tu.

86

Table 7. Frame structure parameters for a maximum speed of 500 km/h.

Centre frequency (GHz) 2 10 28 30 38 73Frame duration (ms) 10 10 10 10 10 10Subframe duration (ms) 1 1 1 1 1 1Cyclic prefix length TCP1 (µs) 5.2 5.2 5.2 5.2 5.2 5.2Cyclic prefix length TCP2 (µs) 4.7 4.7 4.7 4.7 4.7 4.7Doppler frequency FD (kHz) 0.93 4.63 13 14 17.6 33.8Sub-carrier spacing ∆ f (kHz) 15 20 60 60 80 160Length of symbol Tu (µs) 66.7 50 16.7 16.7 12.5 6.25Number of symbol Nsym 7 9 23 23 29 46Resource block (RB) size (kHz) 180 240 720 720 960 1920Resource elements per RB 84 108 276 276 348 552Number of pilot symbols per RB 4 6 14 14 16 26

Although, at higher frequency bands, the delay spread is generally small, we assume thisinformation is unknown and assume the worst case channel condition (at least using theexisting state-of-the-art CP length) is used for the system design. However, the optimalselection of cyclic prefix for different frequency bands can be obtained from statisticalknowledge of the delay spread for different propagating channels.

Furthermore, it is important to examine the fundamental relationship and impact ofbandwidth and noise power on the achievable SNR/SINR to show the limits of increasingthe channel bandwidth and provide an appropriate lower bound on the SNR/SINR toachieve a successful transmission of a signal. If we assume an SISO link, the achievabledata rate over the link is limited by the capacity of the link, which is a function of thebandwidth B and the SNR Γ

C = B× log2(1+Γ). (38)

On the other hand, the SNR is a function of the received power Pr and the noise powerPn. The noise power is proportional to the bandwidth given as

Pn = k0×T0×NF×B, (39)

where T0 is the ambient temperature, NF is the noise figure, and k0 is Boltzmann’sconstant. From (38), we know that increasing the bandwidth is a straightforward wayto improve the achievable data rate. However, the noise power also increases with anincrease in bandwidth. Hence, with a fixed transmit power, a significant increase inbandwidth will lead to a significantly low SNR. A large increase in the transmit powercould be seen as a solution. However, a large amount of transmit power is inappropriatedue to the high power consumption, power amplifier requirements, heat-dissipation

87

problems, and emission regulations [87]. Similarly, when multiple cells are considered,the presence of inter-cell interference I needs to be considered and the achievable datarate is expressed as

Racv = B× log2

(1+

Pr

N + I

)= B× log2 (1+Γ) (40)

where the SNR is replaced with signal to noise plus interference ratio (SINR).

1 2 M

BS

velocity info feedbackBeamform

selection

Beamform

selection

balise

H

Fig. 18. Single-cell train scenario [94] ©2017.

For a given fixed bandwidth, an improved data rate can be achieved by ensuring highSINR. The achieved SINR on a given BS-UE link is dependent on the carrier frequencyfc and the distance d between the serving BS and UE, expressed as

Γ =Pr

N + I=

Pbt ×Gb

t ×10−PL( fc,d)

10

N +∑I(Pb′t ×Gb′

t ×δ b′ ×10−PL( fc,dI )

10 )(41)

88

where Pbt and Pb′

t are the transmit powers from the serving BS and interfering BSsrespectively, Gb

t and Gb′t are the transmit antenna gains at the serving BS and interfering

BSs respectively, δ b′ corresponds to the load factor estimated at the interfering BSsand PL(.) is the pathloss, which can be obtained using any of the pathloss models in[127–132]. The effect of the carrier frequency on the minimum SNR/SINR for reliabletransmission is evaluated in Section 4.4.1.

4.2 System model

We consider a train communication scenario with an emphasis on the BS-to-trainlink as shown in Fig. 18. The train has multiple carriages, each equipped with asingle moving relay node (MRN). The number of MRNs is denoted by M. The BS isequipped with Nt transmit antennas with Nr f RF chains so that Nt is a multiple of Nr f .Each MRN is equipped with an external antenna array with Nr receive antennas. Letpm,n = [xm,n,ym, j,ht ]

T ∈R3 and q = [0,0,hb]T ∈R3 be the locations of the mth MRN on

the HST and the BS. The downlink received signal vector for a single MRN yc,i ∈ CNr

at the cth sub-carrier for the ith RF chain is given as

yc,i = Hc,imn,isn,i +n (42)

where Hc,i ∈ CNr×Nti is the channel matrix of the ith RF chain between the serving BSand the MRN. The beamforming weight is given as mn,i ∈ CNti , where the subscript n

is the index from a beamforming set M and sn,i ∈ C denotes the corresponding datasymbol for the ith RF chain. The additive complex white Gaussian noise vector is definedas n∼ CN (0,N0INr) with zero mean and N0 variance. The RF chains are implementedsuch that the channel can be expressed as Hc = [Hc,1, ...,Hc,Nr f ]. Hence, the receivedsignal to noise ratio (SNR) at the cth sub-carrier can be written as

Γc =|wH

n Hcmn|2

‖wn‖2N0, (43)

where wn ∈ CNr denotes the receive filter for the MRN.

4.3 Beamforming for higher operating frequencies on railwaynetworks

In this section, an HFB beamforming codebook and selection algorithm are developed tomaximize the achievable throughput on an HST without requiring any training overhead.

89

The problem formulation and general structure of the codebook is presented in Section4.3.1. The benefit of large array gain and its impact on the beamwidth is examined inSection 4.3.2, while the impact of the link distance on the angle of arrival/departure andhow to deal with its effect are considered in Section 4.3.3. In Section 4.3.4, the impactof velocity estimation errors on the angle of arrival/departure is analyzed and the HSTbeamforming selection algorithm is presented in Section 4.3.5.

4.3.1 Problem formulation

In order to maximize the received SNR, a joint transmit/receive beamforming is proposed,i.e.,

maximizemn,wn

Γc

subject to mn ∈M

wn ∈W

(44)

where M and W are the sets of transmit and receive vectors. With channel knowledge atthe BS and HST, the optimum beamformers can be solved by singular value decom-position (SVD) if Nt = Nr f . However, in a high mobility scenario, full knowledge ofthe channel is not feasible. With no knowledge of the channel, the optimal weightscan be solved by an exhaustive search through the sets M and W . However, in areal-time sensitive scenario like the HST, the selected weights will be outdated due tofeedback delay. Taking into account the characteristics of the railway environment andthe directionality of HFB propagation, the AoA and AoD of the LOS propagation pathcan be estimated based on prior knowledge of the HST position and velocity. Hence, theoptimal beamforming weight mn and receive filter wn can be aligned with the estimatedAoA and AoD so that the steering vectors are expressed as

mn = at(θAoDlmax

,φ AoDlmax

)

wn = ar(θAoAlmax

,φ AoDlmax

)(45)

where θ AoAlmax

(φ AoAlmax

) and θ AoDlmax

(φ AoDlmax

) are the azimuth (elevation) angle of arrival anddeparture of the strongest multipath component, respectively. The sets M and W arecomposed of steering vectors from which the optimum vectors are selected. These sets

90

are expressed as

M = at(θ1,φ1), ....,at(θNb ,φNb)

W = ar(θ1,φ1), ....,ar(θNb ,φNb)

|θn,φ ∈ [0,π),n = 1,2, ...,Nb,

(46)

where the codebook based sets M and W are finite and limited by Nb. We propose thatthe beam-form selection criteria are tied to the velocity feedback as shown in Fig. 18.The velocity feedback can be obtained using balises installed at roughly regular intervalsalong the track as shown in Fig. 18.

A balise is an electronic beacon or transponder placed between the rails of a railwayas a part of an automatic train protection system. It is currently an integral part of theEuropean train control system (ETCS) that gives the exact location of a train. Theyare also used in the Chinese train control system. Originally, the balises are safetyequipment that transmit information about the location of the balise, the geometry of thetrack region to the next balise and speed restrictions. The balise is capable of receivinginformation from the train, but currently this feature is rarely used. We propose that thefixed information (e.g., the geometry of the track) on each balise is made available at theBSs that cover the region of the track where these balises are placed, and once the HSTcrosses over a balise, the balise is activated and receives the HST speed information. Thebalises usually come in pairs, which can be used to determine the direction of the HST(velocity) and this information along with the balise index number is fed back to the BS.

4.3.2 Relationship between array gain and beamwidth

For HFB channels, the channel matrix can be mainly characterized by the transmit andreceive array response vectors so that

Hc =

√NtNr

L

L

∑l=1

αlar(θAoAl ,φ AoA

l )at(θAoDl ,φ AoD

l )H (47)

where αl is the amplitude of the channel gain of the lth multipath component. ar(.) andat(.) are the receive and transmit array response vectors respectively, while θ AoA

l (φ AoAl )

and θ AoDl (φ AoD

l ) represent the azimuth (elevation) angle of arrival and departure of thelth multipath component, respectively. The array response vector can be expressed as

a(θ) = [1,e j(k∆sin(θ)+δ ), ...,e j((N−1)k∆sin(θ)+δ )]T (48)

91

where k = 2π/λ , λ is the carrier wavelength, ∆ is the antenna spacing and δ is thephase shift from element to element. For the HST scenario, the far field conditions areassumed to be met, since Nt ·∆t ·λ dm,n and Nr ·∆r ·λ dm,n can easily be satisfied.Therefore, the array response vector can form beams in the direction of the phase shift δ

with the beam gain expressed as

Bg = ∑a(θ) = e j(N−12 ψ) sin(N

2 ψ)

sin(ψ/2), with ψ = k∆sin(θ)+δ . (49)

Assuming omnidirectional antenna elements, the maximum array gain is N and thebeamwidth is 1.78/N centering at the steering angle, which is achieved when ψ = 0(see proof in Appendix 2). The array gain is beneficial when there is a beam alignmentbetween the transmit and receive antenna arrays. The direction of the beam can beelectronically controlled by phase shifters to follow the movement of the HST withδ =−k∆sin(θ0), where θ0 is the steering angle. The beam alignment problem can besolved if the steering angle is always aligned with the angle of arrival. It is intuitive thatincreasing the number of antenna elements in the array will lead to an increased arraygain. However, an increase in the array gain will lead to a narrower beamwidth makingthe beam alignment problem more challenging, as reflected in Table 8.

Table 8. Relationship between Bg and Bw [92] ©2017.

Size of antenna array (N) Beam gain (Bg) Beamwidth (Bw)

8 8 12.8°

16 16 6.4°

32 32 3.2°

64 64 1.6°

4.3.3 Impact of BS-HST link distance on the angle of arrival

For the steering angle to be aligned with the angle of arrival while still maintaining higharray gain, precise and real-time main lobe alignment between the BS and the HST isrequired, as the HST moves at high speed. Knowledge of the rate of change of the angleof arrival w.r.t. the horizontal coordinate distance xm,n and the geometry of the trackis required to help achieve a compromise between the array gain and the beamwidth.Following the propagation scenario D2 in [95], where the minimum perpendiculardistance between the track and the BS d is 50 m with a BS height hb and MRN height ht

92

of 30 m and 5 m respectively, the rate of change of the angle of arrival is obtained asshown in Fig. 19. We assume a straight track with BSs in the intervals of 1000 m.

-500 -400 -300 -200 -100 0 100 200 300 400 500

Horizontal coordinate distance (m)

20

30

40

50

60

70

80

90

Ang

le o

f arr

ival

(de

gree

s)

Fig. 19. Rate of change of angle of arrival w.r.t. BS-HST separation distance [92] ©2017 IEEE.

Fig. 19 shows how the angle of arrival changes with the increasing distance between theBS and the HST. The change in the angle of arrival varies rapidly when the HST is closeto the BS within the range of approximately 200 m. As the separation distance increasesbeyond 200 m, the rate of change of the angle of arrival reduces. This implies thatwith prior knowledge of the location and velocity of the HST, electronic steering of themain-lobe with a high array gain can be achieved with precision when the HST is fartheraway from the BS and more challenging at close range. To solve the beam alignmentproblem at short range, the beamwidth can be widened to accommodate the rapid changein AoA. One way to widen the beamwidth is the use of the deactivation approach,where some antenna elements within the antenna array are deactivated. However, withper-antenna power constraint the maximum total power will be limited. An alternative tothe deactivation approach is to split the large antenna array into sub-arrays with Nts

antennas in each sub-array. The sub-arrays can be implemented so that the channel

93

can be expressed as H =√

Nt NrL ∑

Ll=1 αlaS

r (aSt )

H with the array response vector definedas aS = vec([a1S(θ AoA

1∗,1 ), ...,aSS(θ AoA

1∗,S )])T , where the angles used for the sub-arrays are

carefully selected to widen the beamwidth by a factor of S2. (see proof in Appendix 2)

4.3.4 Impact of velocity estimation error on the angle of arrival

In practice, the prior knowledge of the HST’s position and velocity might slightlydeviate from the real time position and velocity due to environmental factors such asthe direction and strength of the wind. This deviation can sometimes be critical andlead to beam misalignment. Following the geometry in Fig. 20, let pm,n = [Px,Py,Pz]

T

be the initial position of the train w.r.t. the BS. The estimated train position pm,n atany time instance can be obtained in terms of the velocity information, i.e., ‖pm,n‖=v(τn− τn−1)+‖pm,n−1‖. pm,n−1 represents the initial tracked position (n−1) of thetrain and τn denotes the time required to get to the nth position on the track based on theestimated velocity v.

12M

hb

ht

d

xm,n u0

1,nAoD

1,nAoA

1,nAoD

1,nAoA

12

Ntr

Fig. 20. HST scenario with angle of arrival based beamforming [92] ©2017 IEEE.

We assume the deviation in the known velocity ge of the HST follows a normaldistribution with the mean µ and variance σ2 so that the v = v+ge. The deviation in theAoA as a result of v is given as

θerror = tan−1(

∆xm,n

d +∆d

). (50)

94

The relative displacement along the perpendicular distance between the BS and theHST is denoted by ∆d, which can occur as a result of wind strength. The symbol∆xm,n = |xm,n− xm,n| is the relative horizontal distance displacement caused by the errorin the velocity of the HST. The relative displacement ∆d is assumed negligible due to thelarge distance d. Hence, the outage probability will depend on the relative displacement∆xm,n and the size of the beamwidth Bw. Therefore, the outage probability is defined as

Pout = Probθerror > αBw. (51)

Here, the outage probability is defined as the event when the angle of misalignment islarger than a fraction α of the beamwidth of the antenna array.

M =

a1t : a11

t (θ1,1,φ1,1), ....a11t (θNb,1,φNtr ,1), level 1

a2t :a12

t (θ1,1,φ1,1), ....a12t (θNb,1,φNtr ,1)

a22t (θ1,2,φ1,2), ....a22

t (θNb,2,φNtr ,2)

, level 2

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,

. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ,

aSt :

a1St (θ1,1,φ1,1), ....a1S

t (θNb,1,φNtr ,1)

. . . . . . . . . . . . . . . . . . . . . . . . . . . ,

aSSt (θ1,S,φ1,S), ....aSS

t (θNb,S,φNtr ,S)

, level S

|θn ∈ [0,π),n = 1,2, ...,Nb.

|φ j ∈ [π/2,π), j = 1,2, ...,Ntr.

(52)

4.3.5 High speed train beam selection scheme

The beam selection is based on the anticipated AoD and AoA of the dominant path,which is calculated from the prior information of the rail environment. This informationincludes the perpendicular BS-to-track distance d, the geometry of the track, theheight of the BS hb, the height of the HST ht , the velocity v and initial position ofthe HST pm,n. Based on these, the angular domain can be determined by casting anormal three-dimensional positioning calculation for a single dimension. Given theproblem formulation in (9), the sets of transmit M and receive W beamforming vectors,which form the codebook are generated by evenly sampling the angular domain in asequential order with a total interval of Nb on a given track. With the incorporation ofthe sub-array approach discussed in Section 4.3.3, the codebook is designed to consist

95

of S levels corresponding to the number of sub-arrays. Each level contains beamformingvectors with a fixed beamwidth and corresponding array gain. The codebook for thebeamforming vectors is given in (52)

To always ensure beam alignment with the movement of the HST, we proposeAlgorithm 7, which sequentially selects the beamforming weights from M at the endof a count-down-timer (CDT ). The CDT is configured based on the prior knowledgeof the initial HST position and velocity. Using Figure 20 as a reference, pm,n =

[(x1,n−2u(m−1)),d,ht ]T , where 2u is the length of a carriage. The algorithm selects a

codebook level ast from (52) based on pm,n. The codebook level selection is observed

at each transmission time interval (TTI) as the HST position changes. The estimatedvelocity is also updated at each TTI to minimize the error in the configured CDT. Theelapsed time of the CDT triggers the BS to select the next beamforming weights mm

n+i

and the process continues until the last beamforming weights mmNb

are selected, whichthen triggers a handover process to the next BS. The distance between the BS and themth MRN at the nth beam can be described as

dm,n = ‖pm,n−qm,n‖= x2m,n +(hb−ht)

2 +d212 . (53)

The elevation angles of arrival and departure are same across the track and are defined as

φAoAm,n = sin−1

(d2 + x2m,1

12

dm,1

)= sin−1

(d2 + x2m,n

12

dm,n

)φ

AoDm,n = π−φ

AoAm,1 = π−φ

AoAm,n . (54)

On the other hand, the azimuth angles are also a function of the horizontal coordinatedistance xm,n, which varies rapidly due to the speed of the train. The relationshipbetween the azimuth angles and the horizontal coordinate distance can be expressed as

θAoAm,n =

π

2− sin−1

(d

dm,n

)θ

AoDm,n = π−θ

AoAm,n . (55)

4.4 Performance evaluation

In this section, the performance evaluation on the suitability of the use of HFBs for HSTcommunication and the proposed beamforming selection scheme are presented. TheHFB feasibility is evaluated in Section 4.4.1 and the proposed beamforming selectionscheme is evaluated in Section 4.4.2.

96

Algorithm 7 Proposed Beam Selection Scheme.Input:

1: BS retrieves:2: Initial HST position pm,n = [xm,n,ym, j,zm,r]

T

3: Estimated velocity of the HST v = v+ge

4: Geometry of track and have MInitialization:

5: Select ast to use based on pm,n

6: for m≤M do . M- number of MRNs on the HST7: BS uses mm

n based on initial location pm,n

8: end forIteration:

9: for each TTI do . TTI- transmission time interval10: for n < Nb do11: τn = ‖pm,n+1−pm,n‖/v

12: end for13: i← 014: while i < Nb do . Nb- number of beam direction in codebook15: CDT = τn+i

16: Start CDT . count down in nanoseconds17: if CDT == 0 then18: Select as

t to use based on new pm,n

19: for m≤M do20: BS uses next mm

n+i based on new location pm,n+i

21: end for22: i← i+123: Continue24: end if25: end while26: Update v

27: end for

97

4.4.1 HFB feasibility evaluation

Required pathloss gains and SNR/SINR lower bound

First, we briefly compare the effect on the pathloss observed for HFB signals withmicrowave frequency signals using conventional and higher frequency proposed pathlossmodels. To this end, we focus on the amount of gain required by HFB signals tomaintain the same pathloss as for the microwave signals. Knowledge on the amount ofgain required is particularly useful in the HST network, since it is important to be ablemaintain the macro cell size currently used by microwave bands to avoid extremelyfrequent handovers as the HST moves across multiple cells at high speed.

Fig. 21. Additional required gain w.r.t. 2 GHz [94] ©2017.

The required gain needed at carrier frequencies of 10, 28, 30, 38 and 73 GHz to achievethe same pathloss with a carrier frequency of 2 GHz is shown in Fig. 21 with a fixed linkdistance of 900 m. The bar chart shows the additional gain required at HFBs to maintainthe same pathloss as at an operating frequency of 2 GHz comparing the four differentpathloss models, free space, IEEE 802.16, 3GPP LOS Rma and CI pathloss models.

98

For each of the pathloss models examined, it can be seen in Fig. 21 that the higher thecarrier frequency, the larger the additional gain required to maintain same pathloss asthe 2 GHz carrier frequency. It can also be observed that the required gains for thefree space, 3GPP LOS Rma, and CI pathloss models are similar to the required gainsapproximately 14, 23, 24, 26, and 32 dB for carrier frequencies 10, 28, 30, 38, and 73GHz, respectively. This shows that the pathloss models’ dependencies on the carrierfrequency are the same and since these three pathloss models exhibit a single slopecurve, the dependency on distance is also similar. The modified IEE 802.16 pathlossmodel has higher required gains compared to the other pathloss models as a result of adifference in the dependency of the carrier frequency.

Fig. 22. Modified IEEE 802.16: Required gain w.r.t. 2 GHz [94] ©2017.

Furthermore, the modified IEEE 802.16 pathloss model exhibits a varying dependencyon distance w.r.t. the carrier frequency as seen in Fig. 22. The figure shows a similarrequired gain for a range of distances, which changes w.r.t the carrier frequency. As thecarrier frequency increases, the range of distance which follows the free space pathlosspattern becomes shorter.

99

Table 9. Obtaining the bandwidth efficiency factor for different frequency bands.

Frequency (GHz) 2 10 28 30 38 73

Ncp (%) 6.68 8.56 21.72 21.72 27.36 43.34

Npilot (%) 4.76 5.56 5.07 5.07 4.6 4.7

NL1/L2 (%) 28.5 28.5 28.5 28.5 28.5 28.5

NACLR (%) 10 10 10 10 10 10

α 0.57 0.52 0.48 0.48 0.45 0.35

The required gain for HFBs to maintain the same pathloss as the 2 GHz frequency bandare closely in the same range irrespective of the pathloss model used. If for a given linkdistance, the required gain compensation for the HFBs are applied to achieve the sameSNR across the frequency bands, the same target rate can be achieved at low mobility.However, due to the high mobility of the HST and the sensitivity to Doppler and delayspread by the HFBs, large frame errors can occur. Hence a minimum SNR/SINR forreliable communication at a given target rate needs to be defined taking into account thehigh mobility of the HST and the sensitivity to Doppler and delay spread by HFBs.

The minimum SNR/SINR required for successful demodulation of a transmittedsignal is obtained using (34). The bandwidth efficiency factor given in Table 9 wasderived from (35) and the 5G new radio parameters in Table 7.The SNR efficiency factor was obtained by extrapolation from the lookup table mappingbetween CQI and modulation scheme in [133]. Fig. 23 shows the minimum requiredSNR/SINR ratios to achieve successful transmission with different spectral efficiencies.The figure shows that to achieve a 0.2 bps/Hz spectral efficiency, the observed SINRmust be greater than -5 dB for the 2 GHz curve. For HFBs, the minimum SNR/SINRtargets are higher, which show that the higher the carrier frequency the higher thesensitivity to high mobility. The figure also shows that with the same SNR/SINR for thedifferent frequencies, which can be achieved with the additional gain from Fig. 21, theachievable spectral efficiency will vary w.r.t. frequency band. For example, with anSNR/SINR of 5 dB, the achievable spectral efficiency at 10 GHz is 1.2 bps/Hz and at 73GHz, the achievable spectral efficiency is 0.42 bps/Hz.

SINR at cell-edge for different operating frequencies

We consider an interference limited scenario with the aim of evaluating the impact ofinter-cell-interference (ICeI) at different operating frequency bands.

100

Fig. 23. Lower bound on SINR for a given spectral efficiency [94] ©2017.

The achievable SINRs at the cell-edge are compared for the different frequency bandsand pathloss models. A downlink transmission with only large scale fading is assumedusing a 19 tri-sector cell hexagonal layout. The inter-site-distance (ISD) was set to 1730m with a frequency reuse pattern of 1-3-1, where all the adjacent cells use the samefrequency set in order to ensure ICeI. We also assumed a load factor of 1, which is theratio between the used bandwidth and the available bandwidth for the interfering links.

The results in Fig. 24, show that the achievable SINR at the cell-edge decreasesas the bandwidth increases, since the noise power is a function of the bandwidth. Ingeneral, the results show a small reduced difference in the achievable SINR for HFBscompared to the 2 GHz carrier frequency except for the modified IEEE 802.16 pathlossmodel. The effect of higher operating frequencies on the SINR are significant whenconsidering the IEEE 802.16 pathloss model, which is as a result of the varying increasein pathloss w.r.t distance and frequency as reflected in Fig. 22. Considering the threeother pathloss models, there is a minimum of 14 dB additional pathloss compared to the2 GHz frequency band.

101

(a) Bandwidth = 100 MHz

(b) Bandwidth = 200 MHz

(c) Bandwidth = 500 MHz

Fig. 24. Achievable SINR at cell-edge for different operating frequencies at different band-width [94] ©2017.

102

Table 10. Simulation parameters [92] ©2017 IEEE.

Parameters Values

Carrier frequency 28 GHz

Number of channel realizations 100

BS antenna spacing 0.5 wavelength

Length of each carriage 30 m

Number of BS/MRN antennas 8, 16, 32

Angle of arrival/departure Uniform distribution (−π/2,π)

Transmit power Pt 46 dBm

However, in an interference limited scenario as observed in Fig. 24c, the maximumreduction in achievable SINR at the cell edge compared to the 2 GHz frequency bandshows about 6 dB loss for a bandwidth of 500 MHz. This is as a result of the fact that athigher operating frequencies, the interfering paths also experience increased pathlosses.Hence, in an interference limited network, the reduced interference in the networklayout reduces the effect of the increased pathloss on the desired path. Note that thespectral efficiency for a given SINR will vary according to Fig. 23.

4.4.2 Proposed beamforming scheme evaluation

In this subsection, we provide the simulation results to evaluate the performance of theproposed beamforming selection scheme with perfect AoA estimation and comparewith the ideal SVD technique. Then we take into account the effect of BS-HST linkdistance and velocity estimation errors in examining the performance of the proposedbeamforming selection scheme. We consider a BS with Nr f RF chains and Nt transmitantennas serving the HST. Each carriage in the HST is equipped with an MRN withNr receive antennas. The number of sub-carriers is set to C = 1000 with a sub-carrierspacing of 500 kHz for the 28 GHz carrier frequency. The channel model used isobtained from a statistical spatial channel model (SSCM) for HFB LOS communicationlinks [134]. It is based on extensive propagation measurements carried out on the 28 and73 GHz bands. In each simulation, the performance is averaged over 100 independentchannel realizations. The main simulation parameters are listed in Table 10.

Fig. 25 shows the data rates achieved for different SNR values with the SVDbeamforming scheme and the proposed beamforming selection scheme (PBSS) formulti-stream transmission with an 8×8, 16×16 and 32×32 HFB LOS SSCM channel.

103

Fig. 25. Performance comparison between SVD and PBSS for different antenna configura-tions [92] ©2017 IEEE.

The results show a significant increase in spectral efficiency as the antenna arrayincreases from 8 to 32. This is intuitive because as the number of antennas increases, thenumber of spatial degrees of freedom also increases. However, the number of spatialdegrees of freedom is significantly small in relation to the size of the antenna array. Thisphenomenon is reflected in Fig. 25 where the rate performance of the multi-stream PBSSdisplays a close performance to the multi-stream SVD for all antenna configurationsdespite having a small number of RF chain Nr f compared to the size of the antenna array.At an SNR of 15 dB, the performance gap in Fig. 25 show an achieved spectral efficiencydifference of 0.56, 1.65 and 2 b/s/Hz for 8, 16, and 32 antenna arrays, respectively. Notethat Nr f was set to 4 for all antenna configurations except for the 32×32 configurationfor which the value was set to 8.

In Fig. 26, the misalignment of the beams is described by the outage probability as afunction of the velocity error variance σ2.

104

Fig. 26. Outage probability for different antenna configurations [92] ©2017 IEEE.

Fig. 26 shows the outage probability for different antenna array sizes at a horizontalcoordinate distance of xm,n = 200 m and a velocity of v = 500 km/h. The figure showsthat with larger antenna arrays, the sensitivity to velocity estimation errors becomeshigher leading to an increased chance of beam misalignment. For example, at σ2 = 2.5transmission is in outage approximately 25 % of the time with an array size of 8 andin outage 100 % of the time with an array size of 16, 32 and 64. Fig. 27 shows theaverage rate achieved at different BS-HST link distances for an antenna array size of 16.In the figure, four different average rates at each specific distance are shown, where"Perfect velocity estimation" and "Imperfect velocity estimation (IVE)" correspond tothe application of Algorithm 7 assuming a perfect velocity estimation and applyingAlgorithm 7 in the presence of velocity error with a variance σ2 = 0.2, respectively.

The impact of velocity feedback errors are more significant at shorter link distances.To reduce the impact of velocity feedback errors, the deactivation approach and sub-arrayapproach is applied with Algorithm 7.

105

50 m 100 m 200 m 300 m 400 m 500 m

BS-HST link distance

0

1

2

3

4

5

6

7

Rate

(b/s

/Hz)

Imperfect velocity estimation (IVE)

IVE with deactivation approach

IVE with sub-array approach

Perfect velocity estimation

Fig. 27. Average achievable rate at different link BS-HST link distance [92] ©2017 IEEE.

The number of sub-arrays used for the sub-array approach is S = 4, while the antennasused for each link distance for the deactivation approach is given in Table 11. Thesub-array approach in Fig. 27 shows a close performance to the Perfect velocity

estimation since there is a significant reduction in the outage probability as a result ofthe beamwidth expansion by a factor of S2.

Table 11. Active antennas used for deactivation approach [92] ©2017 IEEE.

Link distance (m) 50 100 200 300 400 500

Antennas used (Au) 4 8 12 12 16 16

The deactivation approach in Fig. 27 shows a substantial improvement in the rateperformance compared to the Imperfect velocity estimation. However, the deactivationapproach provides a worse performance compared to the sub-array approach because

106

the total transmit power is limited by the number of antennas used and the beamwidth isonly expanded by a factor of Nt/Au. Hence, in the deactivation approach the probabilityfor transmission to be in outage is higher than in the subarray approach.

4.5 Summary

In the development of 5G networks for HST communications, HFBs will play animportant role in providing high data rates for train passengers. Hence, in this chapter,we examined the feasibility of using HFBs with reference to the 2 GHz band for HSTnetworks. We modified the OFDM frame structure so that the characteristics of HFBsand the high velocity of the HST are taken into consideration. With the modified framestructure, we identified the limits for different HFBs on the minimum SNR/SINR for agiven target spectral efficiency needed to achieve successful and reliable communication.Based on the OFDM frame structure, the symbol length is shortened with an increase inthe carrier frequency, thereby increasing the sensitivity to both ISI and ICI. A fixedcyclic prefix length was used to compensate for the effect of both ISI and ICI, but at thecost of consuming a significant amount of bandwidth, particularly evident at the 73GHz frequency band. We also showed that irrespective of the pathloss model used, thepathloss dependencies on frequency and distance are similar when evaluated in terms ofthe required gain by HFBs to achieve the same performance as microwave bands. Theevaluation on different pathloss models was motivated by the fact that there are manyexisting and ongoing campaign efforts towards 5G pathloss modelling [135] and it willbe important to understand the impact of the choice of pathloss model used.

An HFB timer-based beamforming selection algorithm for HSTs was also proposedin this chapter, which exploits the characteristics of the railway environment and thevelocity feedback of the HST. We developed a sequentially ordered and hierarchicalcodebook based on the array response vectors and tied to the LOS AoA and AoDbetween the BS-HST link. We showed that array gains can be achieved with largerantenna arrays, but this leads to narrower beamwidth and in turn increases the chance ofmisalignment between the transmit and receive beams. We also investigated the impactof velocity feedback errors with different error variances on the outage probability, andtwo approaches were examined for mitigating against the impact of velocity feedbackerrors. The performance of the proposed algorithm was evaluated and the results showedan achievable spectral efficiency comparable to the ideal SVD scheme with a reducedperformance gap of less than 2 b/s/Hz. The results provided in this chapter demonstrates

107

the potential of the use of HFBs for HST networks. However, a maximum carrierfrequency of 38 GHz is suggested for HSTs due to the increased sensitivity to Dopplershift, ISI and ICI at HFBs.

108

5 Conclusion

This thesis has investigated and developed a number of schemes to improve the qualityof service (QoS) of mobile users accessing the wireless communication network at highspeed. In particular, the high speed train (HST) with a dual-hop network architecturewas the focus of this thesis. This thesis aims were to provide practical-orientedschemes which exploit the characteristics of the railway environment to improve thecommunication throughput on an HST. Due to the practical nature of these schemes, astraightforward consideration for 5G and beyond can be achieved.

Chapter 1 highlighted the motivation and assumptions made for these studies.Chapter 1 also described the challenges of wireless communications at high speed and areview of prior studies was discussed.

In Chapter 2, practical realizable transmission schemes for HST wireless communi-cations were developed. The results in Chapter 2 showed that the key point for improvingthe throughput performance in an HST scenario is to use sufficiently large antenna arraysat the BS and on the train. The results also demonstrated that precoding is somewhatirrelevant, but transmission rank adaptation via exhaustive search type of process ishighly beneficial. However, rank adaptation requires a lot of computation, especiallywhen using large antenna arrays. The key insight in reducing the computational loadof the system was to simplify the exhaustive search process by comparing only themost relevant combinations, i.e., focusing on choosing the right number of transmitantennas rather than going through all the possible combinations. It was shown that theproposed low complexity algorithms including spatial multiplexing with simplifiedantenna selection provided performance which was almost comparable with that ofthe exhaustive search scheme, but with a notably reduced computational load. Largeantenna arrays at the transmitting and receiving sides, with simple spatial multiplexingand antenna selection transmission scheme appear to be an appropriate solution tosignificantly improve the backhaul link throughput in a practical HST communicationscenario. These simplified algorithms also fit well into demanding real-time scenariossuch as for high speed moving receivers, where the time taken to select an appropriatetransmission scheme is significantly reduced as a result of the reduced computationalload.

109

Chapter 3 considered an HST in an existing cellular network infrastructure, wherenormal macro users have to share resources with HSTs that are temporarily located in thesame cell due to the speed of the HST. Taking into account the different groups of mobileterminals (MTs) the scheduling approach will optimize the achievable throughput andimprove fairness among each group of MTs. The results in Chapter 3 showed that thetype of scheduling approach used can significantly impact the throughput performancefor both the normal ground users and onboard users in the HST. It was also shown thatapart from the additional cost that can be avoided from using dedicated base stationsfor the HST, further reductions in capital expenditure can be achieved with a trade-offbetween the number of MRNs and the size of antenna array without compromising onthe throughput performance.

Chapter 4 considered higher frequency bands (HFBs) for transmission on HSTs.First the feasibility of the use of HFBs was examined and then a transmit beamformingselection algorithm was developed based on the peculiar propagation characteristics ofHFBs and the railway environment. The results in Chapter 4, showed that without anysignificant deviation from state of the art frame structure, the use of HFBs for HSTcommunications is conveniently feasible until a maximum carrier frequency of 38 GHz.The proposed time-based beamforming selection algorithm showed that throughputperformance can be closely maintained when compared to the ideal SVD scheme withperfect alignment between the transmit and receive beams. The impact of velocityfeedback errors on the margin of misalignment was examined and a sub array methodwas found to be effective in widening the beam, thereby mitigating against the impact ofvelocity feedback errors.

In this thesis, the primary target was to maximize the throughput which is an essentialcomponent in the future 5G high mobility wireless network. However, the end-to-enddelays in a network are also a key parameter influencing the QoS of high mobility usersaccessing the network. For instance, achieving a high data rate with an overly largeend-to-end delay will not improve the users’ experience. Therefore, minimizing theend-to-end delay should be a focus for future work. Introducing direct communicationbetween the trains in conjunction with the base station-to-train link is a concept that candrastically reduce the end-to-end delay. However, with a high speed of up to 500 km/h,two trains on the same track may not perform direct communications adequately due tosevere pathloss. The pathloss is caused by the minimum required safety distance ofabout 8 km if absolute block system is implemented on the rail and about 3 km if the

110

dynamic block system is in place. A multi-hop concept can be introduced in whichtrains on other tracks in close proximity can initiate communication.

With the rapid increase in the number of connected devices with diverse capabilitiesand requirements, 5G networks are expected to provide better scalability for handlingthe increasing number of connected devices. These diverse connected devices create achallenge for optimized resource sharing. Hence, resource allocation methodology andscheduling approaches for multiple groups of devices could be considered for futurework. The concept of network slicing could be introduced, where the different groups ofconnected devices can be defined.

The feasibility of the use of higher frequency for HST wireless communicationsbased on OFDM frame structure was examined. However, future studies could examineother multiple access schemes such as the non- and quasi-orthogonal or filter bankmulticarrier (FBMC) multiple access methods for comparison.

Other types of vehicles could be taken into account, where the assumptions onchannel estimation will have to be addressed differently. For example, the LOScomponent may not exist in this new propagation environment. Hence, the high speedof the vehicle could lead to significant Doppler spread. Doppler shifts can easily becompensated for if the speed and direction of movement is known, but Doppler spreadcan be more problematic.

111

112

References

1. Cisco (2017). Cisco visual networking index: Global mobile data traffic forecastupdate, 2016 − 2021. White paper. URI: https://www.cisco.com/c/en/us/

solutions/collateral/service-provider/visual-networking-index-vni/

mobile-white-paper-c11-520862.html. Cited: 2017/10/19.2. Liang HW & Hwang YH (2016) Mobile phone use behaviors and postures on public

transportation systems. PloS one 11(2): 1–12.3. Andrews JG, Buzzi S, Choi W, Hanly SV, Lozano A, Soong AC & Zhang JC (2014) What

will 5G be? IEEE J. Select. Areas Commun. 32(6): 1065–1082.4. Osseiran A, Boccardi F, Braun V, Kusume K, Marsch P, Maternia M, Queseth O, Schellmann

M, Schotten H, Taoka H et al. (2014) Scenarios for 5G mobile and wireless communications:the vision of the METIS project. IEEE Commun. Mag. 52(5): 26–35.

5. 5G use-cases. Ericsson. URI: http://www.ericsson.com/assets/local/news/2015/7/5g-use-cases.pdf. Cited: 2017/09/30.

6. Fan P (2014) Advances in broadband wireless communications under high-mobilityscenarios. Chinese Science Bulletin 59(35): 4974–4975.

7. Dat PT, Kanno A, Yamamoto N & Kawanishi T (2015) WDM RoF-MMW and linearlylocated distributed antenna system for future high-speed railway communications. IEEECommun. Mag. 53(10): 86–94.

8. Zhu X, Chen S, Hu H, Su X & Shi Y (2013) TDD-based mobile communication solutionsfor high-speed railway scenarios. IEEE Wireless Commun. 20(6): 22–29.

9. Wu J & Fan P (2016) A survey on high mobility wireless communications: Challenges,opportunities and solutions. IEEE Access 4: 450–476.

10. Samsung (2015). 5G vision. White paper. URI: http://www.samsung.com/global/business-images/insights/2015/Samsung-5G-Vision-0.pdf. Cited: 2017/09/30.

11. Giordani M, Mezzavilla M & Zorzi M (2016) Initial access in 5G mmWave cellular networks.IEEE Commun. Mag. 54(11): 40–47.

12. Gelabert X, Legg P & Qvarfordt C (2013) Small cell densification requirements in highcapacity future cellular networks. In: Proc. IEEE Int. Conf. Commun. Workshop, pp.1112–1116.

13. Hasna MO & Alouini MS (2003) End-to-end performance of transmission systems withrelays over Rayleigh-fading channels. IEEE Trans. Wireless Commun. 2(6): 1126–1131.

14. Pennanen H, Tolli A & Latva-aho M (2014) Multi-cell beamforming with decentralizedcoordination in cognitive and cellular networks. IEEE Trans. on Signal Processing 62(2):295–308.

15. Ai B, Cheng X, Kurner T, Zhong ZD, Guan K, He RS, Xiong L, Matolak D, Michelson D &Briso-Rodriguez C (2014) Challenges toward wireless communications for high-speedrailway. IEEE Trans. Intell. Transp. Syst. 15(5): 2143–2158.

16. 3GPP (2009). LTE; evolved universal terrestrial radio access(E-UTRA); SI applicationprotocol, 3GPP Rep. TS 36.413 V8.6.1. Technical Specification.

17. Lopez-Perez D, Guvenc I & Chu X (2012) Mobility management challenges in 3GPPheterogeneous networks. IEEE Commun. Mag. 50(12).

113

18. Dimou K, Wang M, Yang Y, Kazmi M, Larmo A, Pettersson J, Muller W & Timner Y(2009) Handover within 3GPP LTE: design principles and performance. In: Proc. IEEE Veh.Technol. Conf., pp. 1–5.

19. Lin CC, Sandrasegaran K, Ramli H, Basukala R, Patachaianand R, Chen L & Afrin TS(2011) Optimization of handover algorithms in 3GPP long term evolution system. In: Proc.IEEE Modeling, Simulation and Applied Optimization Conf.(ICMSAO), pp. 1–5.

20. Tanghe E, Joseph W, Verloock L & Martens L (2008) Evaluation of vehicle penetration lossat wireless communication frequencies. IEEE Trans. Veh. Technol. 57(4): 2036–2041.

21. Berens F, Dunger H, Czarnecki S, Bock T, Reuter R, Zeisberg S, Weber J & Guasch JF(2007) UWB car attenuation measurements. In: Proc. Mobile and Wireless Commun.Summit, pp. 1–5.

22. Virk UT, Haneda K, Kolmonen VM, Vainikainen P & Kaipainen Y (2014) Characterizationof vehicle penetration loss at wireless communication frequencies. In: Proc. European Conf.Antennas Propag. (EuCAP), pp. 234–238.

23. Zhou Y, Pan Z, Hu J, Shi J & Mo X (2011) Broadband wireless communications on highspeed trains. In: Proc. IEEE Wireless and Optical Commun. Conf.(WOCC), pp. 1–6.

24. Wang J, Zhu H & Gomes NJ (2012) Distributed antenna systems for mobile communicationsin high speed trains. IEEE J. Sel. Areas Commun. 30(4): 675–683.

25. 3GPP (2016). 3rd generation partnership project; technical specification group radio accessnetwork; study on scenarios and requirements for next generation access technologies 3GPPRep. TR 38.913 V14.0.0. Technical Specification.

26. Bianchi G, Blefari-Melazzi N, Grazioni E, Salsano S & Sangregorio V (2003) Internetaccess on fast trains: 802.11-based on-board wireless distribution network alternatives. In:Proc. Mobile and Wireless Commun. Summit, pp. 15–18.

27. Van Phan V, Horneman K, Yu L & Vihriälä J (2010) Providing enhanced cellular coveragein public transportation with smart relay systems. In: Proc. IEEE Veh. Networking Conf.,pp. 301–308.

28. Liu Z & Fan P (2014) An effective handover scheme based on antenna selection inground–train distributed antenna systems. IEEE Trans. Veh. Technol. 63(7): 3342–3350.

29. Tian L, Li J, Huang Y, Shi J & Zhou J (2012) Seamless dual-link handover scheme inbroadband wireless communication systems for high-speed rail. IEEE J. Sel. Areas Commun.30(4): 708–718.

30. Luo W, Zhang R & Fang X (2012) A CoMP soft handover scheme for LTE systems in highspeed railway. EURASIP J. Wireless Commun. and Networking 2012(1): 1–9.

31. Bölcskei H (2006) MIMO-OFDM wireless systems: basics, perspectives, and challenges.IEEE Trans. Wireless Commun. 13(4): 31–37.

32. Abed-Meraim K, Moulines E & Loubaton P (1997) Prediction error method for second-orderblind identification. IEEE Trans. Signal Processing 45(3): 694–705.

33. Wang H, Lin Y & Chen B (2003) Data-efficient blind OFDM channel estimation usingreceiver diversity. IEEE Trans. Signal Processing 51(10): 2613–2623.

34. Muquet B, De Courville M & Duhamel P (2002) Subspace-based blind and semi-blindchannel estimation for OFDM systems. IEEE Trans. Signal Processing 50(7): 1699–1712.

35. Li C & Roy S (2003) Subspace-based blind channel estimation for OFDM by exploitingvirtual carriers. IEEE Trans. Wireless Commun. 2(1): 141–150.

36. Khalighi MA & Boutros JJ (2006) Semi-blind channel estimation using the EM algorithm initerative MIMO APP detectors. IEEE Trans. Wireless Commun. 5(11).

114

37. Li X & Fan P (2012) Modified clustered comb pilot-aided fast time-varying channelestimation for OFDM system. J. of Modern Transportation 20(4): 220–226.

38. Cannizzaro RC, Banelli P & Leus G (2006) Adaptive channel estimation for OFDM systemswith Doppler spread. In: Proc. IEEE Signal Processing Advances Wireless Commun., pp.1–5.

39. Tang Z, Cannizzaro RC, Leus G & Banelli P (2007) Pilot-assisted time-varying channelestimation for OFDM systems. IEEE Trans. Signal Processing 55(5): 2226–2238.

40. Cheng P, Chen Z, Rui Y, Guo YJ, Gui L, Tao M & Zhang Q (2013) Channel estimation forOFDM systems over doubly selective channels: A distributed compressive sensing basedapproach. IEEE Trans. Commun. 61(10): 4173–4185.

41. Ren X, Tao M & Chen W (2016) Compressed channel estimation with position-based ICIelimination for high-mobility SIMO-OFDM systems. IEEE Trans. Veh. Technol. 65(8):6204–6216.

42. Simon EP, Ros L, Hijazi H, Fang J, Gaillot DP & Berbineau M (2011) Joint carrier frequencyoffset and fast time-varying channel estimation for MIMO-OFDM systems. IEEE Trans.Veh. Technol. 60(3): 955–965.

43. Hui B, Kim J, Chung HS, Kim IG & Lee H (2016) Efficient Doppler mitigation forhigh-speed rail communications. In: Proc. Int. Conf. Advanced Commun. Technol. (ICACT),pp. 634–638.

44. Klotsche R, Wünstel K & Banniza TR (2010). Doppler compensation control for radiotransmission. US Patent 7,653,347.

45. Li J & Zhao Y (2012) Radio environment map-based cognitive Doppler spread compensationalgorithms for high-speed rail broadband mobile communications. EURASIP JournalWireless Commun. and Networking 2012(1): 1–18.

46. Luoto P, Rikkinen K, Kinnunen P, Karjalainen J, Pajukoski K, Hulkkonen J & Latva-aho M(2017) Configurable 5G air interface for high speed scenario. In: Proc. European Conf.Networks and Commun., pp. 1–5.

47. Assaad M (2009) Reduction of the feedback delay impact on the performance of schedulingin OFDMA systems. In: Proc. IEEE Veh. Technol. Conf., pp. 1–4.

48. Basukala R, Ramli H, Sandrasegaran K & Chen L (2010) Impact of CQI feedback rate/delayon scheduling video streaming services in LTE downlink. In: Proc. IEEE Int. Conf. Commun.Technol., pp. 1349–1352.

49. Dahlman E, Parkvall S & Skold J (2013) 4G: LTE/LTE-advanced for mobile broadband.Academic press.

50. Dong Z, Fan P, Panayirci E & Mathiopoulos PT (2012) Effect of power and rate adaptation onthe spectral efficiency of MQAM/OFDM system under very fast fading channels. EURASIPJ. Wireless Commun. and Networking 2012(1): 208.

51. Sternad M, Grieger M, Apelfröjd R, Svensson T, Aronsson D & Martinez AB (2012) Usingpredictor antennas for long-range prediction of fast fading for moving relays. In: Proc.IEEE Wireless Commun. and Networking Conf. Workshops, pp. 253–257.

52. Laiyemo A & Scott S (2016). Controlling transmission. Google Patents. URI: https://encrypted.google.com/patents/EP3007368A1$?$cl$=$und. EP 3007368 A1.

53. Bolcskei H, Gesbert D & Paulraj AJ (2002) On the capacity of OFDM-based spatialmultiplexing systems. IEEE Trans. Commun. 50(2): 225–234.

54. Luo W, Fang X, Cheng M & Zhao Y (2013) Efficient multiple-group multiple-antenna(MGMA) scheme for high-speed railway viaducts. IEEE Trans. Veh. Technol. 62(6):

115

2558–2569.55. Cheng X & Fang X (2014) Principal component analysis based multiplexing solution

for MIMO systems in high-speed railway. In: Proc. High Mobility Wireless Commun.(HMWC), pp. 48–52.

56. Zhang JY, Tan ZH & Wang HB (2011) Optimum capacity of MIMO systems for high-speedrailway with spare antenna array. In: Proc. IEEE Veh. Technol. Conf., pp. 1–4.

57. Sun N & Wu J (2014) Maximizing spectral efficiency for high mobility systems withimperfect channel state information. IEEE Trans. Wireless Commun. 13(3): 1462–1470.

58. Valentin S & Wild T (2010) Studying the sum capacity of mobile multiuser diversity systemswith feedback errors and delay. In: Proc. IEEE Veh. Technol. Conf., pp. 1–5.

59. Yilmaz A & Kucur O (2014) Performances of transmit antenna selection, receive antennaselection, and maximal-ratio-combining-based hybrid techniques in the presence of feedbackerrors. IEEE Trans. Veh. Technol. 63(4): 1976–1982.

60. Li L, Vorobyov SA & Gershman AB (2009) Transmit antenna selection based strategies inMISO communication systems with low-rate channel state feedback. IEEE Trans. WirelessCommun. 8(4): 1660–1666.

61. Mun C (2006) Transmit-antenna selection for spatial multiplexing with ordered successiveinterference cancellation. IEEE Trans. Commun. 54(3): 423–429.

62. Gore DA, Heath Jr RW & Paulraj AJ (2002) Transmit selection in spatial multiplexingsystems. IEEE Commun. Lett. 6(11): 491–493.

63. Heath Jr RW, Sandhu S & Paulraj A (2001) Antenna selection for spatial multiplexingsystems with linear receivers. IEEE Commun. Lett. 5(4): 142–144.

64. Laiyemo A, Pennanen H, Pirinen P & Latva-aho M (2017) Transmission strategies forthroughput maximization in high speed train communications: From theoretical study topractical algorithms. IEEE Trans. Veh. Technol. 66(4): 2997–3011.

65. Scott S, Leinonen J, Pirinen P, Vihriälä J, Van Phan V & Latva-aho M (2013) A cooperativemoving relay node system deployment in a high speed train. In: Proc. IEEE Veh. Technol.Conf., pp. 1–5.

66. Laiyemo AO, Pirinen P & Latva-aho M (2014) Alternative to dynamic rank transmission forLTE mobile relay node system. In: Proc. European Wireless Conf., pp. 1–6.

67. Laiyemo AO, Pirinen P, Latva-aho M, Vihriala J & Van Phan V (2013) Impact of LTEprecoding for fixed and adaptive rank transmission in moving relay node system. In: Proc.Int. Conf. ITS Telecommun., pp. 250–254.

68. Fokum DT & Frost VS (2010) A survey on methods for broadband internet access on trains.IEEE Commun. Surveys Tuts. 12(2): 171–185.

69. Tingting G & Bin S (2010) A high-speed railway mobile communication system based onLTE. In: Proc. IEEE Int. Conf. Electron. and Inform. Eng., volume 1, pp. 414–417.

70. Sui Y, Vihriälä J, Papadogiannis A, Sternad M, Yang W & Svensson T (2013) Moving cells:a promising solution to boost performance for vehicular users. IEEE Commun. Mag. 51(6):62–68.

71. Wang LC & Lin WJ (2004) Throughput and fairness enhancement for OFDMA broadbandwireless access systems using the maximum C/I scheduling. In: Proc. IEEE Veh. Technol.Conf., volume 7, pp. 4696–4700.

72. Anchun W, Liang X, Shidong Z, Xibin X & Yan Y (2003) Dynamic resource managementin the fourth generation wireless systems. In: Proc. Int. Conf. Commun. Technol., volume 2,pp. 1095–1098.

116

73. Ruangchaijatupon N & Ji Y (2008) Simple proportional fairness scheduling for OFDMAframe-based wireless systems. In: Proc. IEEE Wireless Commun. and Networking Conf.,pp. 1593–1597.

74. Zhu HJ & Hafez RH (2007) Scheduling schemes for multimedia service in wireless OFDMsystems. IEEE Wireless Commun. 14(5).

75. Sun Q, Tian H, Dong K, Wang S & Zhang P (2008) A novel resource allocation algorithm formultiuser downlink MIMO-OFDMA. In: Proc. IEEE Wireless Commun. and NetworkingConf., pp. 1844–1848.

76. Chung WC, Chang CJ & Wang LC (2012) An intelligent priority resource allocation schemefor LTE-A downlink systems. IEEE Wireless Commun. Lett. 1(3): 241–244.

77. Challa N & Cam H (2004) Cost-aware downlink scheduling of shared channels for cellularnetworks with relays. In: Proc. Int. Conf. Perf., Computing, and Commun., pp. 793–798.

78. Viswanathan H & Mukherjee S (2005) Performance of cellular networks with relays andcentralized scheduling. IEEE Trans. Wireless Commun. 4(5): 2318–2328.

79. Huang L, Rong M, Wang L, Xue Y & Schulz E (2007) Resource scheduling forOFDMA/TDD based relay enhanced cellular networks. In: Proc. IEEE Wireless Commun.and Networking Conf., pp. 1544–1548.

80. Zhu H (2012) Radio resource allocation for OFDMA systems in high speed environments.IEEE J. Select. Areas Commun. 30(4): 748–759.

81. Zhao Y, Li X, Li Y & Ji H (2013) Resource allocation for high-speed railway downlinkMIMO-OFDM system using quantum-behaved particle swarm optimization. In: Proc. IEEEInt. Conf. Commun., pp. 2343–2347.

82. Laiyemo AO, Pennanen H, Pirinen P & Latva-aho M (2016) Resource scheduling approachfor LTE-A based network incorporating a moving relay node system equipped train. In:Proc. Global Wireless Summit, pp. 1–5.

83. Guan K, Li G, Kürner T, Molisch AF, Peng B, He R, Hui B, Kim J & Zhong Z (2017) Onmillimeter wave and THz mobile radio channel for smart rail mobility. IEEE Trans. Veh.Technol. 66(7): 5658–5674.

84. Ai B, Guan K, Rupp M, Kurner T, Cheng X, Yin XF, Wang Q, Ma GY, Li Y, Xiong L et al.(2015) Future railway services-oriented mobile communications network. IEEE Commun.Mag. 53(10): 78–85.

85. Roh W, Seol JY, Park J, Lee B, Lee J, Kim Y, Cho J, Cheun K & Aryanfar F (2014)Millimeter-wave beamforming as an enabling technology for 5G cellular communications:Theoretical feasibility and prototype results. IEEE Commun. Mag. 52(2): 106–113.

86. Rangan S, Rappaport TS & Erkip E (2014) Millimeter-wave cellular wireless networks:Potentials and challenges. Proc. of the IEEE 102(3): 366–385.

87. Maltsev A, Sadri A, Pudeyev A & Bolotin I (2016) Highly directional steerable anten-nas: High-gain antennas supporting user mobility or beam switching for reconfigurablebackhauling. IEEE Veh. Technol. Mag. 11(1): 32–39.

88. Cui Y, Fang X & Yan L (2016) Hybrid spatial modulation beamforming for mmWaverailway communication systems. IEEE Trans. Veh. Technol. 65(12): 9597–9606.

89. Kim J & Molisch AF (2013) Enabling Gigabit services for IEEE 802.11ad-capable high-speed train networks. In: Proc. IEEE Radio and Wireless Symp. (RWS), pp. 145–147.

90. Va V, Zhang X & Heath RW (2015) Beam switching for millimeter wave communication tosupport high speed trains. In: Proc. IEEE Veh. Technol. Conf., pp. 1–5.

117

91. Song H, Fang X & Fang Y (2016) Millimeter-wave network architectures for futurehigh-speed railway communications: Challenges and solutions. IEEE Wireless Commun.23(6): 114–122.

92. Laiyemo AO, Luoto P, Pirinen P & Latva-aho M (2017) Higher frequency band beamformingscheme for high speed train. In: Proc. IEEE Veh. Networking Conf., pp. 1–6.

93. Laiyemo AO, Pennanen H, Pirinen P & Latva-aho M (2016) Effective deployment ofcooperative moving relay nodes in a high speed train. In: Proc. Wireless Days (WD), pp.1–6.

94. Laiyemo AO, Luoto P, Pirinen P & Latva-aho M (2017) Feasibility studies on the use ofhigher frequency bands and beamforming selection scheme for high speed train communica-tion. Wireless Commun. and Mobile Computing 2017.

95. Meinilä J, Kyösti P, Jämsä T & Hentilä L (2009) WINNER II channel models. RadioTechnologies and Concepts for IMT-Advanced pp. 39–92.

96. Kaleva J, Tölli A & Juntti M (2012) Weighted sum rate maximization for interferingbroadcast channel via successive convex approximation. In: Proc. IEEE Global Commun.Conf., pp. 3838–3843.

97. Shi Q, Razaviyayn M, Luo ZQ & He C (2011) An iteratively weighted MMSE approach todistributed sum-utility maximization for a MIMO interfering broadcast channel. IEEETrans. Signal Process. 59(9): 4331–4340.

98. Tse D & Viswanath P (2005) Fundamentals of Wireless Communication. CambridgeUniversity Press.

99. Love DJ, Heath RW, Santipach W & Honig ML (2004) What is the value of limited feedbackfor MIMO channels? IEEE Commun. Mag. 42(10): 54–59.

100. Hochwald BM, Marzetta TL, Richardson TJ, Sweldens W & Urbanke R (2000) Systematicdesign of unitary space-time constellations. IEEE Trans. Inf. Theory 46(6): 1962–1973.

101. Dabak AG, Lin C, Onggosanusi EN & Varadarajan B (2011). Codebook and pre-coderselection for closed-loop MIMO. US Patent 7,949,064.

102. Lee J, Han JK & Zhang J (2009) MIMO technologies in 3GPP LTE and LTE-Advanced.EURASIP J. Wireless Commun. and Networking 2009.

103. Roessler J (2015). LTE-Advanced (3GPP Rel. 12) technology introduction. White paper.104. Weichselberger W, Herdin M, Özcelik H & Bonek E (2006) A stochastic MIMO channel

model with joint correlation of both link ends. IEEE Trans. Wireless Commun. 5(1):90–100.

105. 3GPP (2010) 3rd generation partnership project; Technical specification group radio accessnetwork; Evolved universal terrestrial radio access (E-UTRA); Further advancements forE-UTRA physical layer aspects," 3GPP, Rep. TR 36.814 V9.0.0. Technical report.

106. Kermoal JP, Schumacher L, Pedersen KI, Mogensen PE & Frederiksen F (2002) A stochasticMIMO radio channel model with experimental validation. IEEE J. Sel. Areas Commun.20(6): 1211–1226.

107. 3GPP (2012) 3rd generation partnership project; Technical specification group radio accessnetwork; Evolved universal terrestrial radio access (E-UTRA); Physical layer procedures,"3GPP, Rep. TR 36.213 V10.6.0. Technical report.

108. He X, Niu K, He Z & Lin J (2007) Link layer abstraction in MIMO-OFDM system. In:Proc. IEEE Int. Workshop on Cross Layer Design (IWCLD), pp. 41–44.

109. Series M (2009) Guidelines for evaluation of radio interface technologies for IMT-Advanced.Technical report, ITU.

118

110. Tenorio S, Spence P, Garriga B, López J, García A & Arranz M (2010) 3G HSPA forbroadband communications with high speed vehicles. In: Proc. IEEE Veh. Technol. Conf.,pp. 1–5.

111. Chen L, Huang Y, Xie F, Gao Y, Chu L, He H, Li Y, Liang F & Yuan Y (2013) Mobile relayin LTE-advanced systems. IEEE Commun. Mag. 51(11): 144–151.

112. He R, Ai B, Wang G, Guan K, Zhong Z, Molisch AF, Briso-Rodriguez C & Oestges CP(2016) High-speed railway communications: From GSM-R to LTE-R. IEEE Veh. Technol.Mag. 11(3): 49–58.

113. Rappaport TS, Gutierrez F, Ben-Dor E, Murdock JN, Qiao Y & Tamir JI (2013) Broadbandmillimeter-wave propagation measurements and models using adaptive-beam antennas foroutdoor urban cellular communications. IEEE Trans. Antennas Propag. 61(4): 1850–1859.

114. Zhou T, Tao C, Salous S, Liu L & Tan Z (2015) Channel sounding for high-speed railwaycommunication systems. IEEE Commun. Mag. 53(10): 70–77.

115. He R, Zhong Z & Ai B (2010) Path loss measurements and analysis for high-speed railwayviaduct scene. In: Proc. Int. Wireless Commun. and Mobile Computing Conf., pp. 266–270.

116. He R, Zhong Z, Ai B, Wang G, Ding J & Molisch AF (2013) Measurements and analysis ofpropagation channels in high-speed railway viaducts. IEEE Trans. Wireless Commun. 12(2):794–805.

117. He R, Zhong Z, Ai B & Ding J (2012) Measurements and analysis of short-term fadingbehavior for high-speed rail viaduct scenario. In: Proc. IEEE Int. Conf. Commun., pp.4563–4567.

118. He R, Zhong Z, Ai B, Ding J, Yang Y & Molisch AF (2013) Short-term fading behavior inhigh-speed railway cutting scenario: measurements, analysis, and statistical models. IEEETrans. Antennas Propag. 61(4): 2209–2222.

119. Luan F, Zhang Y, Xiao L, Zhou C & Zhou S (2013) Fading characteristics of wirelesschannel on high-speed railway in hilly terrain scenario. Int. J. Antennas and Propag. 2013.

120. Aikio P, Gruber R & Vainikainen P (1998) Wideband radio channel measurements for traintunnels. In: Proc. IEEE Veh. Technol. Conf., volume 1, pp. 460–464.

121. 3GPP (2016). 3rd generation partnership project; Technical specification group radio accessNetwork; Study on scenarios and requirements for next generation access technologies,3GPP, Rep. TR 38.913 V14.2.0. Technical Specification.

122. Müller MK, Taranetz M & Rupp M (2015) Providing current and future cellular services tohigh speed trains. IEEE Commun. Mag. 53(10): 96–101.

123. research Rysavy (2016). Mobile broadband transformation: LTE to 5G. White paper.124. Mogensen P, Na W, Kovács IZ, Frederiksen F, Pokhariyal A, Pedersen KI, Kolding T, Hugl

K & Kuusela M (2007) LTE capacity compared to the Shannon bound. In: Proc. IEEE Veh.Technol. Conf., pp. 1234–1238.

125. Sesia S, Baker M & Toufik I (2011) LTE-the UMTS long term evolution: from theory topractice. John Wiley & Sons.

126. Zyren J & McCoy W (2007) Overview of the 3GPP long term evolution physical layer.Freescale Semiconductor, Inc., white paper .

127. Goldsmith A (2005) Wireless communications. Cambridge university press.128. Naden M & Hart M (2006) Multihop path loss model (base-to-relay and base-to-mobile).

IEEE 802.16 Session 43.129. IEEE (2007). Multi-hop relay system evaluation methodology (channel model and perfor-

mance metric). IEEE 802.16 Broadband Wireless Access Working Group.

119

130. 3GPP (2016) 3rd generation partnership project; Technical specification group radio accessnetwork; Study on channel model for frequency spectrum above 6 GHz 3GPP, Rep. TR38.900 V14.1.0. Technical report.

131. Rappaport TS, MacCartney GR, Samimi MK & Sun S (2015) Wideband millimeter-wavepropagation measurements and channel models for future wireless communication systemdesign. IEEE Trans. Commun. 63(9): 3029–3056.

132. Sun S, Rappaport TS, Thomas TA, Ghosh A, Nguyen HC, Kovács IZ, Rodriguez I, KoymenO & Partyka A (2016) Investigation of prediction accuracy, sensitivity, and parameterstability of large-scale propagation path loss models for 5G wireless communications. IEEETrans. Veh. Technol. 65(5): 2843–2860.

133. 3GPP (2016). 3rd generation partnership project; Technical specification group radio accessnetwork; Evolved universal terrestrial radio access (E-UTRA); Physical layer procedures3GPP, Rep. TS 36.213 V14.1.0. Technical Specification.

134. Samimi MK & Rappaport TS (2016) 3-D millimeter-wave statistical channel model for 5Gwireless system design. IEEE Trans. Microw. Theory Tech. 64(7): 2207–2225.

135. Haneda K, Tian L, Asplund H, Li J, Wang Y, Steer D, Li C, Balercia T, Lee S, Kim Y et al.(2016) Indoor 5G 3GPP-like channel models for office and shopping mall environments. In:Proc. IEEE Int. Conf. Commun. Workshops, pp. 694–699.

120

Appendix 1 Optimum MRN and receive antennaselection flow chart

#MRN = #Carriages Nr = x

#MRN -1#ant_add = 0

50%-tileCDF Throughput (Ref. - Curr.)<=δ

Nr +1#ant_add ++

#MRN +1 Nr -1

#MRN <= β

#MRN*; Nr*

#ant_add>=α

NO

YES

YES YES

NO

NO

121

122

Appendix 2 Beam analysis

Beam gain Bg

Bg = ∑(a(θ)) =N

∑n=1

exp j(n−1)ψ

(exp jψ Bg−Bg) =N

∑n=1

exp jnψ−N

∑n=1

exp j(n−1)ψ

Bg =exp jNψ−1exp jψ−1

=exp j(N/2)ψ(exp j(N/2)ψ−exp− j(N/2)ψ)

exp j(N/2)ψ(exp j(1/2)ψ−exp− j(1/2)ψ)

Bg = exp j(N−1)ψ/2 sin(N/2)ψsinψ/2

exp j(N−1)ψ/2 accounts for the fact that the physical center of the array is located at(N−1)∆/2 and will therefore be neglected. The maximum gain is achieved when ψ = 0and following the squeeze theorem we can show that

limx→0

sinxx

= 1, therefore, limψ→0

Bg = limψ→0

sin(N/2)ψsinψ/2

= N

Beam width Bw

If we assume ψ/2 is small, we can apply small angle approximation to Bg and normalizeto yield

Bgn =1N

sin(N/2)ψψ/2

The beamwidth is obtained when Bgn = 0.707

sin(N/2)ψ(N/2)ψ

= 0.707, this implies (N/2)ψ =±1.391

ψ = k∆sinθ±+δ =±2.8/N

θ± = sin−1(

1k∆

(±2.8

N−δ

))

123

With half-wavelength spacing ∆ = 0.5λ and an element phase shift δ = 0, we obtain

θ± = sin−1 ±0.89N

and for a large value of N,

small angle approximation holds yielding.

θ± =±0.89

N, Hence

Bw = |θ+−θ−|= 1.78/N

Beam expansion using the sub-array approach

Let S be the number of sub-arrays with Ns antenna elements in each sub-array. ThereforeBg can be rewritten as

Bg =S

∑s=1

Ns

∑n=1

exp j((s−1)Ns+(n−1))ψ

=S

∑s=1

exp j(s−1)Nsψ×Ns

∑n=1

exp j(n−1)ψ

=S

∑s=1

exp j(s−1)Nsψ×BNsg

The first term is the multiplying factor that achieves the maximum value S when ψ = 0and BNs

g is the beam gain for one sub-array. The total beamwidth Bw can be seen asthe union of the beamwidth of each sub-array Bs

w, i.e., Bw = ∪Ss=1BNs

g = S×1.78/Ns =

S2×1.78/N.

124

A C T A U N I V E R S I T A T I S O U L U E N S I S

Book orders:Granum: Virtual book storehttp://granum.uta.fi/granum/

S E R I E S C T E C H N I C A

649. Verrollot, Jordan (2018) Mature supply management as an enabler for rapidproduct development and product portfolio renewal

650. Kemppainen, Anssi (2018) Adaptive methods for autonomous environmentalmodelling

651. Nyländen, Teemu (2018) Application specific programmable processors forreconfigurable self-powered devices

652. Asres, Georgies Alene (2018) Synthesis, characterization and application of WS?nanowire-nanoflake hybrid nanostructures

653. Hajimammadov, Rashad (2018) Plasmonic, electrical and catalytic properties ofone-dimensional copper nanowires : effect of native oxides

654. Barua, Bidushi (2018) Incentivizing user participation in cooperative contentdelivery for wireless networks

655. Pallaspuro, Sakari (2018) On the factors affecting the ductile-brittle transition inas-quenched fully and partially martensitic low-carbon steels

656. Kyösti, Pekka (2018) Radio channel modelling for 5G telecommunication systemevaluation and over the air testing

657. Petäjäjärvi, Juha (2018) Low-power wireless communications in the Internet ofThings – solutions and evaluations

658. Boulkenafet, Zinelabidine (2018) Face presentation attack detection using textureanalysis

659. Kaikkonen, Harri (2018) Supporting rapid product development with agiledevelopment methodologies

660. Tervo, Oskari (2018) Transceiver optimization for energy-efficient multiantennacellular networks

661. Menberu, Meseret Walle (2018) Hydrology of peat-dominated headwatercatchments : Theories and empirical analysis of the impacts of anthropogenicdisturbance

662. Hietava, Anne (2018) Electrical behaviour of submerged arc furnace’s chargematerials

663. Lappalainen, K. Matti (2018) Itämeren rehevöitymisen uudistettu diagnoosi japaradigma

664. Ahmad, Ijaz (2018) Improving software defined cognitive and secure networking

C665etukansi.fm Page 2 Thursday, May 17, 2018 8:45 AM

UNIVERSITY OF OULU P .O. Box 8000 F I -90014 UNIVERSITY OF OULU FINLAND

A C T A U N I V E R S I T A T I S O U L U E N S I S

University Lecturer Tuomo Glumoff

University Lecturer Santeri Palviainen

Postdoctoral research fellow Sanna Taskila

Professor Olli Vuolteenaho

University Lecturer Veli-Matti Ulvinen

Planning Director Pertti Tikkanen

Professor Jari Juga

University Lecturer Anu Soikkeli

Professor Olli Vuolteenaho

Publications Editor Kirsti Nurkkala

ISBN 978-952-62-1956-1 (Paperback)ISBN 978-952-62-1957-8 (PDF)ISSN 0355-3213 (Print)ISSN 1796-2226 (Online)

U N I V E R S I TAT I S O U L U E N S I SACTAC

TECHNICA

U N I V E R S I TAT I S O U L U E N S I SACTAC

TECHNICA

OULU 2018

C 665

Ayotunde Oluwaseun Laiyemo

HIGH SPEED MOVING NETWORKS IN FUTURE WIRELESS SYSTEMS

UNIVERSITY OF OULU GRADUATE SCHOOL;UNIVERSITY OF OULU,FACULTY OF INFORMATION TECHNOLOGY AND ELECTRICAL ENGINEERING;CENTRE FOR WIRELESS COMMUNICATIONS

C 665

AC

TAA

yotunde Oluw

aseun Laiyemo

C665etukansi.fm Page 1 Thursday, May 17, 2018 8:45 AM