Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Enhancements to the long term evolution (LTE) standard for facilitating the Internet of things (IoT) Mysore Balasubramanya, Naveen 2017

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


24-ubc_2017_september_mysorebalasubramanya_naveen.pdf [ 5.41MB ]
JSON: 24-1.0347534.json
JSON-LD: 24-1.0347534-ld.json
RDF/XML (Pretty): 24-1.0347534-rdf.xml
RDF/JSON: 24-1.0347534-rdf.json
Turtle: 24-1.0347534-turtle.txt
N-Triples: 24-1.0347534-rdf-ntriples.txt
Original Record: 24-1.0347534-source.json
Full Text

Full Text

Enhancements to the Long Term Evolution (LTE) Standard forFacilitating the Internet of Things (IoT)byNaveen Mysore BalasubramanyaM.S., University of Colorado, Boulder, USA, 2010B.E., The National Institute of Engineering, Mysore, India, 2005A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctoral Studies(Electrical and Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)May 2017c© Naveen Mysore Balasubramanya, 2017AbstractThe world is evolving towards an Internet of Things (IoT) where a large numberof devices interact to realize different applications that constitute smart electricitygrids, intelligent transportation systems, ubiquitous healthcare solutions, etc. Ma-chine Type Communications (MTC) provide the substrate for the connectivity andservice mechanisms of these devices. Many services associated with the MTC appli-cations such as smart metering and location tracking require the cellular network asthe backbone for communication and are instrumental in driving the growth of theThird Generation Partnership Project (3GPP) Long Term Evolution (LTE)/LTE-Advanced (LTE-A) standards. A substantial number of MTC User Equipment (UE)hosting IoT applications are expected to be low cost, low data rate devices requiringprolonged battery life.In the downlink, the current LTE/LTE-A standards adopt Discontinuous Re-ception (DRX) mechanism for power reduction, which requires the UE to wake upperiodically to check for a paging message from the base station. The LTE/LTE-A standardization activities have identified that intricate paging decode proceduresincrease the energy consumption for low complexity MTC UEs, necessitating en-hancements to the current mechanisms. This encourages us to investigate novelenergy efficient mechanisms for LTE MTC systems. Specifically, we develop DRXwith Quick Sleeping Indication (QSI), which enables the MTC UEs to go back tosleep quickly and save power, when there is no impending page from the base sta-iiAbstracttion. We also design the enhanced Primary Synchronization Signal (ePSS) for fastertiming resynchronization, which can be used as QSI for additional improvements inthe downlink energy efficiency of MTC UEs in low coverage. Further, LTE/LTE-Astandardization activities in the uplink are examining different procedures to reduceUE data retransmissions for improved energy efficiency. To this end, we develop aMaximum Likelihood (ML) based uplink Carrier Frequency Offset (CFO) estima-tion technique for the LTE/LTE-A base station, which is robust and accurate in lowcoverage, enhancing the uplink energy efficiency of MTC UEs.The MTC mechanisms described in this thesis are not only simple to implement,but also require minimal changes to the present LTE/LTE-A standardization frame-work, promoting smooth integration into the current LTE/LTE-A networks.iiiLay SummaryThe Internet is advancing into an extensive, smart platform inter-connecting a varietyof devices leading to an Internet of Things (IoT). A significant portion of the IoT de-vices are expected to be low-cost, low-complexity User Equipment (UE) like sensorsand smart meters. Current wireless communication technologies like the Long TermEvolution (LTE) are defining novel Machine Type Communications (MTC) mecha-nisms to effectively host such devices. In this thesis, we propose enhancements to thepower saving mechanism currently adopted for reception by the LTE MTC UEs andpresent a strategy to reduce UE data retransmissions. Our solutions address batteryoperated UEs, especially those in low network coverage and result in improved en-ergy efficiency, thus increasing their battery life. Moreover, our enhanced mechanismscan be easily integrated into the current LTE platforms, since they require minimalchanges to the present LTE standardization framework.ivPrefaceThe material presented in this thesis is entirely based on research performed bymyself under the supervision of Prof. Lutz Lampe in the Department of Electricaland Computer Engineering (ECE) at the University of British Columbia (UBC),Vancuver, Canada.The co-authors in my publications, Mr. Gustav Vos and Mr. Steve Bennett, fromSierra Wireless Inc.1, Richmond, BC, Canada, have assisted me towards the problemformulation and in determining the relevance of the solutions with respect to theongoing standardization efforts in the Third Generation Partnership Project (3GPP)Long Term Evolution (LTE).Below is a list of publications related to the work presented in this thesis.Publications Related to Chapter 2• Naveen Mysore Balasubramanya, Lutz Lampe, Gustav Vos, and Steve Bennett,“Introducing quick sleeping using the broadcast channel for 3GPP LTE MTC,”IEEE Globecom Workshops, 2014, pp. 606-611.• Naveen Mysore Balasubramanya, Lutz Lampe, Gustav Vos, and Steve Bennett,“DRX with quick sleeping: A novel mechanism for energy-efficient IoT usingLTE/LTE-A,” IEEE Internet of Things Journal (Special Issue on Internet of1https://www.sierrawireless.comvPrefaceThings Over LTE/LTE-A Network: Theory, Methods, and Case Studies), vol.3, no. 3, pp. 398-407, 2016.• Naveen Mysore Balasubramanya, Lutz Lampe and Gustav Vos,“Quick PagingMethod and Apparatus in LTE, US Patent App. 14/799513, filed in 2014.Publications Related to Chapter 3• Naveen Mysore Balasubramanya, Lutz Lampe, Gustav Vos, and Steve Ben-nett, “On Timing Reacquisition and Enhanced Primary Synchronization Signal(ePSS) Design for Energy Efficient 3GPP LTE MTC,” IEEE Transactions onMobile Computing, In print.• Gustav Vos, Naveen Mysore Balasubramanya, Steve Bennett and Lutz Lampe,“Method and System for providing and using Enhanced Primary Synchroniza-tion Signal for LTE,” US Patent App. 15/010, 192, filed in 2015.Publications Related to Chapter 4• Naveen Mysore Balasubramanya, Lutz Lampe, Gustav Vos, and Steve Bennett,“Low SNR Uplink CFO Estimation for Energy Efficient IoT Using LTE,” IEEEAccess (Special Section: The Plethora of Research on the Internet of Things),vol. 4, pp. 3936-3950, 2016.• Naveen Mysore Balasubramanya, Lutz Lampe, Gustav Vos and Steve Ben-nett,“Method and System for Carrier Frequency Offset Estimation in LTEMTC Device Communication,” US Provisional Patent App. 62/307, 327, filedin 2016.viTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xivAcknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xixDedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxi1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Machine Type Communications (MTC) . . . . . . . . . . . . . . . . 41.1.1 Challenges in MTC . . . . . . . . . . . . . . . . . . . . . . . 41.1.2 MTC Solutions . . . . . . . . . . . . . . . . . . . . . . . . . . 61.1.3 Thesis Outline and Major Contributions . . . . . . . . . . . . 121.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14viiTable of Contents1.2.1 DRX and Paging Mechanism in LTE/LTE-A . . . . . . . . . 141.2.2 Narrow-Band Internet of Things (NB-IoT) in LTE/LTE-A . . 191.3 Literature Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221.3.1 Prior Work on Power Saving Mode and DRX Analysis . . . . 221.3.2 Prior Work on Energy Efficient M2M Uplink Mechanisms . . 262 Discontinuous Reception (DRX) with Quick Sleeping . . . . . . . 302.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302.2 Details of the QSI Mechanism . . . . . . . . . . . . . . . . . . . . . . 312.2.1 QSI Working Mechanism . . . . . . . . . . . . . . . . . . . . 322.2.2 DRX and Paging with QSI . . . . . . . . . . . . . . . . . . . 352.3 Quick Sleeping Solutions for MTC UEs Without CE . . . . . . . . . 362.3.1 Quick Sleeping Solutions Using the PBCH . . . . . . . . . . . 372.3.2 QSI on PSS/SSS . . . . . . . . . . . . . . . . . . . . . . . . . 432.4 Quick Sleeping Solution for MTC UEs With CE . . . . . . . . . . . 452.5 Energy Consumption and Computational Complexity Analysis . . . 472.5.1 Energy Consumption Analysis . . . . . . . . . . . . . . . . . 472.5.2 Computational Complexity Analysis . . . . . . . . . . . . . . 512.6 Simulation Results and Analysis . . . . . . . . . . . . . . . . . . . . 532.6.1 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . 532.6.2 Energy Efficiency Results . . . . . . . . . . . . . . . . . . . . 602.6.3 Computational Complexity Results . . . . . . . . . . . . . . . 612.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 623 Enhanced Primary Synchronization Signal (ePSS) . . . . . . . . . 643.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 643.2 UE Timing Accuracy and Legacy Timing Acquisition Algorithms . . 65viiiTable of Contents3.2.1 Importance of UE Timing Accuracy . . . . . . . . . . . . . . 693.2.2 Timing Reacquisition Using CP Autocorrelation . . . . . . . 693.2.3 Timing Reacquisition Using Synchronization Signal Detection 713.3 Technicalities of the ePSS . . . . . . . . . . . . . . . . . . . . . . . . 723.3.1 ePSS Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 723.3.2 ePSS Allocation . . . . . . . . . . . . . . . . . . . . . . . . . 773.3.3 ePSS as Quick Sleeping Indication (QSI) . . . . . . . . . . . . 783.4 Performance Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 803.4.1 Reacquisition Performance Analysis . . . . . . . . . . . . . . 813.4.2 Energy Efficiency Analysis . . . . . . . . . . . . . . . . . . . 823.4.3 Battery Lifetime Improvement . . . . . . . . . . . . . . . . . 863.4.4 Analysis of ePSS Transmission Overhead . . . . . . . . . . . 893.4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 913.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 934 Low SNR Uplink CFO Estimation . . . . . . . . . . . . . . . . . . . 954.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 954.2 Effect of CFO Estimation for NB-IoT Uplink . . . . . . . . . . . . . 974.2.1 Energy Consumption Model . . . . . . . . . . . . . . . . . . . 984.2.2 Numerical Results . . . . . . . . . . . . . . . . . . . . . . . . 984.3 Conventional CFO Estimation Techniques . . . . . . . . . . . . . . . 1014.3.1 CP Autocorrelation . . . . . . . . . . . . . . . . . . . . . . . 1024.3.2 Symbol Repetition . . . . . . . . . . . . . . . . . . . . . . . . 1034.4 ML Based CFO estimation . . . . . . . . . . . . . . . . . . . . . . . 1054.4.1 ML Based CFO Estimation Using Repeated Data . . . . . . . 1064.4.2 ML Based CFO Estimation Using the DMRS . . . . . . . . . 108ixTable of Contents4.4.3 Modified Conventional CFO Estimation Scheme for DMRS . 1094.4.4 ML Based CFO Estimation Using Repeated Data with DMRSCompensation . . . . . . . . . . . . . . . . . . . . . . . . . . 1104.5 Simulation Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1114.5.1 Performance of CFO Estimation Techniques . . . . . . . . . . 1114.5.2 MSE and Crame´r-Rao Bound for the Gaussian Channel . . . 1124.5.3 Results for the EPA Channel . . . . . . . . . . . . . . . . . . 1134.5.4 NB-IoT Large Transport Block Transmission . . . . . . . . . 1194.5.5 Energy Efficiency Analysis . . . . . . . . . . . . . . . . . . . 1214.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1254.6.1 Increased DMRS Density Scheme in LTE/LTE-A Uplink . . . 1264.6.2 Application of ML Based CFO Estimation to Non-LTE Sce-narios . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1274.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1275 Conclusions and Directions for Future Work . . . . . . . . . . . . . 1295.1 Directions for Future Work . . . . . . . . . . . . . . . . . . . . . . . 1335.1.1 Coverage Enhancement for LTE MTC Using Massive MIMO 1335.1.2 Exploring LTE in Unlicensed Bands (LTE-U) for MTC . . . . 1355.2 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . 136Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137AppendicesA Proof of Crame´r-Rao Bound in Eq. (4.13) . . . . . . . . . . . . . . . 148xList of Tables2.1 QSI transmission methods using unused subcarriers on PSS and SSS . 442.2 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 542.3 Performance summary using S10% (without CE) and tAcc (with CE) . 582.4 Reduction in energy consumption for UE with QSI . . . . . . . . . . 602.5 FFT computation reduction for UE with QSI . . . . . . . . . . . . . 623.1 Simulation settings for Figure 3.1. . . . . . . . . . . . . . . . . . . . . 673.2 Rx energy efficiency gain for LC devices in DRX mode using ePSS. . 853.3 Battery lifetime gain for LC devices using ePSS with tDRX = 10.24 sand tDRX = 2621.44 s. . . . . . . . . . . . . . . . . . . . . . . . . . . . 873.4 Network resource overhead due to the ePSS transmission. . . . . . . . 904.1 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 1004.2 Energy efficiency vs. CFO for TBS = 72 bits . . . . . . . . . . . . . . 1004.3 Number of repetitions vs. CFO . . . . . . . . . . . . . . . . . . . . . 1214.4 Energy efficiency gain vs. TBS . . . . . . . . . . . . . . . . . . . . . . 1224.5 Battery lifetime gain vs. TBS for transmitting 10 kilobytes . . . . . . 122xiList of Figures1.1 IoT applications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21.2 LTE radio frame structure showing the different physical channels andpaging transmission. . . . . . . . . . . . . . . . . . . . . . . . . . . . 161.3 Current DRX model in LTE/LTE-A. . . . . . . . . . . . . . . . . . . 181.4 Uplink subframe and PRB in LTE/LTE-A. . . . . . . . . . . . . . . . 192.1 Proposed DRX models. . . . . . . . . . . . . . . . . . . . . . . . . . . 342.2 MIB and PBCH construction. . . . . . . . . . . . . . . . . . . . . . . 382.3 PBCH with QSI transmission . . . . . . . . . . . . . . . . . . . . . . 402.4 Illustration of QSI mechanisms on PBCH. . . . . . . . . . . . . . . . 422.5 QSI transmission mechanism on PDSCH. . . . . . . . . . . . . . . . . 462.6 PBCH BLER performance for AWGN channel. . . . . . . . . . . . . . 552.7 QSI BLER performance for AWGN channel. . . . . . . . . . . . . . . 562.8 Performance results for QSI without CE using the EPA channel model. 572.9 Re-Sync and QSI on PDSCH for CE using the EPA channel model. . 593.1 Illustration of the sensitivity of the PDCCH and the PDSCH decodingto timing offset (TO). . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.2 CP autocorrelation performance. . . . . . . . . . . . . . . . . . . . . 703.3 PSS/SSS detection performance. . . . . . . . . . . . . . . . . . . . . . 713.4 Illustration of ePSS subframe structure for normal and extended CP. 73xiiList of Figures3.5 Illustration of ON time and sleep time for UEs using the legacy andthe ePSS based resynchronization mechanisms. . . . . . . . . . . . . . 803.6 Performance of legacy synchronization signal detection and ePSS de-tection. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.7 Frame Structure and PO Allocation for LC devices using ePSS. . . . 894.1 Illustration of CFO estimation using CP autocorrelation. . . . . . . . 1034.2 Illustration of CFO estimation using symbol repetition. . . . . . . . . 1044.3 MSE vs. SNR and Crame´r-Rao bound for ML based CFO estimationin AWGN. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1124.4 CDF of the estimated CFO error using RV repetitions for legacy LTE/LTE-A uplink. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1144.5 CDF of the estimated CFO error using RV repetitions for MTC LTE/LTE-A uplink. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1154.6 CDF of the estimated CFO error using DMRS only for both legacyand MTC LTE/LTE-A uplink. . . . . . . . . . . . . . . . . . . . . . . 1174.7 CDF of the estimated CFO error using ML estimation and 2x DMRS. 125xiiiList of AbbreviationseNB evolved Node BePSS enhanced Primary Synchronization Signal3GPP Third Generation Partnership Project5G Fifth GenerationARQ Automatic Repeat reQuestAWGN Additive White Gaussian NoiseBLER Block Error RateBPSK Binary Phase Shift KeyingC-RNTI Connected Mode - Radio Network Temporary IdentifierCAT-0 Category-0CAT-1 Category-1CAT-M1 Category-M1CAZAC Complex Amplitude Zero AutoCorrelationCC Convolution CodingCCA Clear Channel AssessmentCDF Cumulative Distribution FunctionCDMA Code Division Multiple AccessCE Coverage EnhancementCFO Carrier Frequency OffsetCP Cyclic PrefixxivList of AbbreviationsCQI Channel Quality IndicationCRC Cyclic Redundancy ChecksumCSI Channel State InformationCTC Convolution Turbo CodeD2D Device-to-DeviceDCI Downlink Control InformationDFT Discrete Fourier TransformDMRS Demodulation Reference SignalsDRX Discontinuous ReceptionEPA Extended Pedestrian AETSI European Telecommunications Standards InstituteFDMA Frequency Division Multiple accessFEC Forward Error CorrectionFFT Fast Fourier TransformGMSK Gaussian Minimum Shift KeyingH-SFN Hyper - System Frame NumberH2H Human-to-HumanICI Inter Carrier InterferenceIEEE Institute of Electrical and Electronics EngineersIFFT Inverse Fast Fourier TransformIMSI International Mobile Subscriber IdentityIoT Internet of ThingsIP Internet ProtocolISM Industrial, Scientific and MedicalLAA Licensed Assisted AccessxvList of AbbreviationsLBT Listen Before TalkLC Low-Cost, Low-Complexity, Low-CoverageLoRa Long RangeLoRaWAN Long Range Wireless Access NetworkLOS Line Of SightLPWA Low Power Wide AreaLRLP Long Range Low PowerLTE Long Term EvolutionLTE-A LTE-AdvancedLTE-U LTE-UnlicensedLTN Low Throughput NetworkM2M Machine-to-MachineMAC Medium Access ControlMCS Modulation and Coding SchemeMIB Master Information BlockMIMO Multi-Input-Multi-OutputML Maximum LikelihoodMSE Mean Squared ErrorMTC Machine Type CommunicationsMU-MIMO Multi-User MIMOMUI Multi-user InterferenceNB-IoT Narrowband Internet of ThingsNLOS Non Line Of SightNR New RadioOFDM Orthogonal Frequency Division MultiplexingxviList of AbbreviationsOFDMA Orthogonal Frequency Division Multiple AccessP-RNTI Paging - Radio Network Temporary IdentifierPBCH Physical Broadcast ChannelPDCCH Physical Downlink Control ChannelPDSCH Physical Downlink Shared ChannelPHICH Physical Hybrid ARQ Indicator ChannelPO Paging OccasionPRACH Physical Random Access ChannelPRB Physical Resource BlockPSS Primary Synchronization SignalQAM Quadrature Amplitude ModulationQoS Quality of ServiceQPSK Quaternary Phase Shift KeyingQSI Quick Sleeping IndicationRACH Random Access ChannelRAN-1 Radio Access network - Layer 1RE Resource ElementRNC Radio Network ControllerRNTI Radio Network Temporary IdentifierRRC Radio Resource ConfigurationRV Redundancy VersionS-TMSI System Architecture Evolution - Temporary MobileSubscriber IdentitySC-FDMA Single Carrier - Frequency Division Multiple AccessSFN System Frame NumberxviiList of AbbreviationsSIB System Information BlockSINR Signal-to-Interference-plus-Noise RatioSISO Single-Input-Single-OutputSNR Signal-to-Noise RatioSON Self Organizing NetworkSSS Secondary Synchronization SignalTBS Transport Block SizeTTI Transmit Time IntervalUE User EquipmentUEGI UE Group IndicatorUEID UE IdentifierUMTS Universal Mobile Telecommunications SystemVCO Voltage Controlled OscillatorVCTCXO Voltage Controlled Temperature Compensated CrystalOscillatorWiFi Wireless FidelityWiMax Wireless Interoperability for Microwave AccessWLAN Wireless Local Area NetworkZC Zadoff-ChuxviiiAcknowledgmentsFirstly, I would like to express my sincere gratitude to my advisor Prof. Lutz Lampefor the continuous support during the course my Ph.D. I thank him for being apatient listener and a constant source of motivation. His knowledge, work ethic andperseverance are truly inspiring and his guidance has helped me in every stage of myresearch and in the writing of this thesis. I could not have imagined having a betteradvisor and mentor for my Ph.D work.Besides my advisor, I would like to thank the rest of my thesis committee: Prof.Victor Leung, Prof. Satish Gopalakrishnan, Prof. Ian Blake and Prof. Cyril Leung,for their time and effort in evaluating my work and providing insightful comments.My sincere thanks also goes to Mr. Gustav Vos and Mr. Steve Bennett fromSierra Wireless Inc., Richmond, BC, Canada, for their valuable inputs and frequentfeedback throughout my Ph.D. research. I would like to thank them for presentingour research in various meetings of the Third Generation Partnership Project (3GPP)Long Term Evolution (LTE) standardization and for their extended support in filingpatents from our research.I am grateful to MITACS, Canada for supporting this research under the MI-TACS Accelerate Fellowship program with the title “New Carrier Types and Ser-vices for Long-term Evolution (LTE) Machine-Type Communication(MTC)”and tothe Natural Science and Engineering Research Council of Canada (NSERC) for theirsupport through the Strategic Project Grant ”Technologies for 5G Wireless Systems:xixAcknowledgmentsSoftware-defined Networks, Massive MIMO, and Full Duplexing”.I thank my fellow labmates for the stimulating discussions and all my friends fortheir encouragement. I would like to thank my family: my parents and my wife, whohave been the biggest support pillars of my life, for lovingly, morally and spirituallyguiding me towards my goals.Last but not the least, I would like to thank my idols - Late Dr. A.P.J AbdulKalam, former president of India and the legendary Indian cricketer Sachin Tendulkar.I have been highly influenced by Dr. Kalam’s books, his lifestyle and words ofwisdom. Sachin Tendulkar constantly met huge expectations from a billion peoplewith great determination and sustained excellence, inspiring me to never give up andkeep chasing my dreams.xxDedicationTo my parents and my wife.xxiChapter 1IntroductionThe Internet is evolving from connecting computers and dedicated terminals to aquintessential medium that can engulf a plethora of “smart” devices like mobilephones, electronic meters, location sensors, etc. The reducing size of silicon on chipand continuously declining price of components have increased the ease of integrationof “smart” sensing and decision-making devices into everyday objects, leading tothe emergence of the Internet of Things (IoT). Diverse applications within the IoTumbrella are not only promising to the consumer, but also appealing to researchersacross various fields. The IoT relies on advancements in different fields such ascommunication technologies, microelectronics, data mining, big data handling, etc.Figure 1.1 summarizes the different applications. A brief account of these applicationsis presented below.• Security and public safety - This category includes applications like homesecurity, building access control, surveillance systems and other public safetymechanisms enabling “smart cities” [1].• Tracking - Numerous applications related to tracking and monitoring are visu-alized over the IoT ranging from monitoring the status of critical infrastructure(nuclear reactors, transport, bridges, etc.) and industrial fleet management [2]to pet tracking for households, wildlife monitoring and cattle tracking in agri-culture [3].1Chapter 1. IntroductionSmart	HealthIoTSecurity	and	Public	SafetySmart	GridTrackingConsumer	devices	and	remote	maintenanceFigure 1.1: IoT applications.• Smart grid - This corresponds to a number of smart applications related tothe distribution and management of power, water, gas and heating. Smartmetering is an appropriate example in this regard [4].• Smart health - Many applications related to automatic monitoring and track-ing of patients, personnel and biomedical devices within hospitals are envisionedunder the “smart health” category [5].• Consumer devices and remote maintenance - Another vital applicationof IoT is in controlling sensors used for vending machine operations, vehiclediagnostics, remote home appliance control, etc. [6].While some of these applications, like driving the sensors that control a nuclear reac-tor are critical in terms of the latency and reliability, others like pet/cattle tracking,smart metering can be more delay tolerant. However, the number of devices support-ing non-critical applications can be massive [7].Wireless communications have been a primary candidate in ensuring the connec-tivity and in satisfying the service requirements of the new, smart devices over the2Chapter 1. IntroductionInternet. The changing landscape of Internet based services has been driving theevolution of wireless communication. As wireless communication technologies stepinto the Fifth Generation (5G), one of the major challenges faced by them is to effec-tively support the diverse requirements of the IoT world. Considering the applicationdiversity and the variety of network scenarios envisioned over the IoT platform, itwould be difficult to have a single solution to answer the requirements of all theapplications.The need to design efficient mechanisms for the IoT devices is being addressed bydifferent standardization committees. The prominent solutions include the Instituteof Electrical and Electronics Engineers (IEEE) 802.15.4 Bluetooth [8, 9] and IEEE802.11 Wireless Local Area Network (WLAN) [10,11], which mainly rely on Device-to-Device (D2D) communication and a distributed network architecture. However,these technologies are limited by low coverage and capacity. The support for IoT isalso being considered by the Third Generation Partnership Project (3GPP) standard-ization committee through the Long Term Evolution (LTE)/LTE-Advanced (LTE-A) standards [12]. The LTE/LTE-A standards use the cellular network for enablinglong-range operation, but enhancements to the current mechanisms are essential foreffectively support the IoT [12].Recently, heterogeneous networks and small cell base-stations [13, 14] are beingdeployed to serve regions where the network traffic is high. Although these solutionsare successful in managing the network load and improving the user throughput, theadditional infrastructure and operational expenses required to realize these solutionsare high. IoT solutions also include Self Organizing Networks (SONs), which havethe ability to improve the network efficiency by adapting, managing and optimizingtheir operations based on the network behaviour [15, 16]. However, SONs require3Chapter 1. Introductioncomplex algorithms and new network equipment to operate efficiently.1.1 Machine Type Communications (MTC)A large number of IoT based services such as automated security systems with mon-itoring and reporting features, pet trackers, agriculture-based applications, etc. areexpected to require wireless network access. Moreover, the devices deployed to real-ize these services are expected to be small sized, cost effective and battery operated.Machine-to-Machine (M2M) or Machine Type Communications (MTC) involves thedefinition, design and development of communication and service mechanisms thatassist in the connectivity of such IoT devices.The MTC mechanisms face a variety of challenges depending on the applica-tion for which the MTC device is being used. These challenges can be completelydifferent from those faced by conventional Human-to-Human (H2H) communicationmechanisms. The key challenges to be addressed in MTC are listed below.1.1.1 Challenges in MTC• Supporting low data rate, delay tolerant operations - While the currentcommunication networks are adept at managing the demands of H2H devices,the IoT scenario requires the network to handle MTC devices with contrastingdemands. For example, an LTE network that is able to provide a good quality ofservice to a high data rate, low latency, videoconferencing application over fewdevices using H2H communication, may not be able to use resources efficientlyfor serving low data rate, delay tolerant MTC devices deployed for smart meterdata reporting.4Chapter 1. Introduction• Handling a massive number of devices - A predominant feature of the IoTis the massive increase in the number of devices requiring network access. Forexample, it is predicted that the cellular IoT has to grow at an average rate of35% annually to reach a potential volume of seven billion units by 2025 [17].Thus, handling massive access, minimizing outage and providing the necessaryQuality of Service (QoS) for different categories of devices would be highlychallenging.• Reducing the energy consumption - MTC devices do not require to beconstantly “connected” or “active”, since their data transmission is not contin-uous and the amount of data to be sent per transmission is small. Furthermore,low-cost and low data rate devices operating with extended battery life (lasting10 or more years) form a substantial part of the IoT equipment. Therefore, itis important to tune the MTC mechanisms so that the energy consumption ofthe MTC devices is reduced [18].• Improving coverage - MTC devices may be located in areas where the net-work coverage is very low, such as basements of buildings, underground parkingfacilities at malls, interiors of hospitals, etc. For example, MTC devices canbe used for medical monitoring, where vital biological parameters of patientssuch as blood pressure, heart rate and body temperature are sensed and ex-changed with a server [19]. The patients have limited mobility within thehospital premises and can be present in a closed, indoor environment where thenetwork requires enhanced coverage to reach them. Due to the restrictions onthe total available power and the maximum power allowed for transmission inthe channel (arising from the spectral mask constraints), the MTC device can-not arbitrarily increase its transmission power to reach the base-station. This5Chapter 1. Introductionresults in a very low operating Signal-to-Noise Ratio (SNR) at the base station.Therefore, it is necessary to design and develop MTC mechanisms that canenhance the performance of devices in low network coverage areas.The contributions of this thesis address energy efficiency and improved coverage as-pects of the IoT MTC devices.1.1.2 MTC SolutionsIn order to drive the IoT, enhanced MTC support is being considered by many cur-rent wireless communication technologies. The traditional short-range communica-tion technologies like Bluetooth and ZigBee [8,9] satisfy the low power requirementsof MTC, but have limited coverage and capacity. While WLAN improves the capac-ity at the expense of power, it falls behind in terms of coverage [10, 11]. Many IoTapplications like smart cities, logistics and health require long range of operation.The range of these technologies can be extended using multi-hop mesh networks,relays and gateways. Small-cell based IoT solutions are prominent candidates in thisregard, however, they come at the expense of additional cost for network infrastruc-ture/maintenance and backhaul provisioning [20]. The cellular technologies providethe desired range, but have higher power consumption. Therefore, it is essential tohave energy efficient, cost-effective solutions addressing the requirements of poten-tially “hard to reach” devices in the network.Although there is no consensus on the percentage of such low coverage, low powerMTC devices in the network, forecasts indicate that they can be significant. LowPower Wide Area (LPWA) M2M connections are expected grow from 20 million toover 860 million by 2020 [21]. Moreover, smart meters are expected to be the largestLPWA application with a projected share of 45% of the total LPWA connections by6Chapter 1. Introduction2020 [21]. Furthermore, low cost and extended coverage are considered to be the keyfeatures of LPWA technologies [20]. Summary of LPWA TechnologiesA hallmark of the LPWA networks is that they provide long range communicationwith reduced power consumption, the desired region of operation for IoT applicationslike smart meters and tracking. It is evident that the LPWA technologies evolvebased on one of the two paths - by increasing the coverage of low power, short-range technologies or by reducing the power of cellular technologies. In this regard,various proprietary LPWA technologies as well as those driven by the standardizationcommittees have emerged.SIGFOX [22] is a patented LPWA technology that independently or by coopera-tion with its network partners offers an end-to-end LPWA connectivity solution. Itoperates in an ultra narrow band, i.e., 100 Hz Sub-GHz Industrial, Scientific andMedical (ISM) band carrier and uses Binary Phase Shift Keying (BPSK) modula-tion. The proprietary SIGFOX base station consists of cognitive software-definedradio, which is connected to an Internet Protocol (IP) based network at the back-end. The use of the ultra narrow band results in decreased noise levels, improvedreceiver sensitivity and low cost antenna design. However, the data rate provided islimited to 100 bps. SIGFOX uses unslotted ALOHA as the Medium Access Control(MAC) layer protocol and does not incorporate any encryption or Forward ErrorCorrection (FEC) for the data packets.Long Range (LoRa) is another prominent, proprietary LPWA technology devel-oped by Semtech Corporation [23]. Similar to SIGFOX, LoRa also operates on theSub-GHz ISM band and uses an unslotted ALOHA based MAC [24]. But, unlike SIG-FOX, it supports data encryption and FEC. For the physical layer, it uses a chirp7Chapter 1. Introductionspread spectrum technique, which results in noise and interference resilient signals.LoRa offers data rates up to 37.5 kbps. Moreover, the upper layers and the systemarchitecture of LoRa are being standardized by LoRa Alliance under Long RangeWireless Access Network (LoRaWAN) specification [25], which supports differentclasses of IoT devices based on latency.Ingenu is an LPWA technology which uses a variant of code division access calledthe random phase multiple access [26], operates in the ISM 2.4 GHz band and offersdata rates up to 68 kbps. Telensa [27] is a candidate similar to SIGFOX. Weightless[28] is another major LPWA technology, operating with three variants - Weightless-W, Weightless-N and Weightless-P. Weightless-W operates in shared TV white-spacesusing 16-ary Quadrature Amplitude Modulation (QAM) and differential-BPSK, pro-viding data rates from 1 kbps to 10 Mbps. Weightless-N operates on ultra narrowband, uses differential-BPSK and can achieve data rates up to 1 kbps. Weightless-Puses a narrow band with Gaussian Minimum Shift Keying (GMSK) and Quater-nary Phase Shift Keying (QPSK) modulation techniques to provide data rates from0.2 kbps to 100 kbps.LPWA has been a regime of interest for distinguished standardization bodies, likethe IEEE, the European Telecommunications Standards Institute (ETSI) and the3GPP. Efforts are in progress for extending the range of IEEE 802.11 standards underthe Long Range Low Power (LRLP) topic by the Task Group AH. Although someuse cases and functional requirements for the LRLP technology have been defined,the standardization and development activities are in the nascent stage [29]. ETSIaims to standardize an LPWA technology called the Low Throughput Network (LTN),which is bidirectional low data rate communication protocol. Specifications have beendefined for the use cases [30], functional architecture [31] and protocols/interfaces [32],8Chapter 1. Introductionwhich include the support for user cooperation and ultra narrow-band operation.SIGFOX, Telensa and LoRa are actively involved in standardizing their technologiesunder ETSI LTN label.While the aforementioned technologies are from the path of extending the range ofshort-range communications, the second path of reducing the power of the long rangecommunications has been the mode of development for the 3GPP cellular standard-ization committee. The leading 3GPP standard in this regard is the LTE/LTE-A,which is expected to occupy more than 75% of the world’s cellular market by 2020 [7].Various improvements to the LTE/LTE-A standard spanning from the network layerto the physical layer are being considered to support the IoT. These include effi-cient routing protocols, parameterization of transmission and reception mechanismsin terms of packet size, modulation schemes, sleep duration, etc.This thesis suggests improvements to the MTC mechanisms in the physical layerof the LTE/LTE-A standards to effectively support the IoT. The motivation behindconsidering physical layer stems from the fact that it is the lower-most layer reflectingthe modifications from the higher layers and the first medium of contact for basebandsignal processing. Improvements to physical layer is essential to leverage the efficiencyof higher layer protocols to host the IoT. MTC in LTE/LTE-A for Facilitating the IoTWith the IoT services emerging as the constitutive driver for the growth of cellularnetwork and MTC assisting the communication mechanisms, 3GPP has initiated thestandardization of MTC from Release 11 of the LTE standard. The major advantagesof using MTC over the LTE network for IoT are that it uses the existing networkinfrastructure to serve the devices, thereby reducing the operational costs and itenables the network operator to harness the higher capacity, coverage and ease of9Chapter 1. Introductionintegration aspects of LTE to serve the devices efficiently. For example, it is shownin [33] that LTE offers significant capacity for smart metering even when low-costdevices are used. The results indicate that approximately 2% of the system resourcesare required to support advanced metering infrastructure in an urban deploymentscenario using LTE. Although LTE provides high capacity, the current LTE/LTE-A networks are designed for efficient H2H communications. In order to utilize thehigh capacity aspects in an efficient manner, the network architecture needs to berevamped to support MTC applications [34–40].In [34], the MTC network architecture is envisioned with simplified physical andmedium access layers that provide sufficient scalability and flexibility for the underly-ing MTC application. An overview of M2M communications supported by LTE/LTE-A is provided in [35] and a grouping-based radio resource management is proposed toachieve the most critical QoS guarantees for MTC devices. The importance of con-gestion control and network overload avoidance for a MTC network architecture andpotential solutions to address these problems are discussed in [36]. The user schedul-ing aspects for 3GPP MTC are presented in [37] and a low complexity schedulerusing a QoS based clustering algorithm is proposed for MTC. The different appli-cation scenarios and the challenges in 3GPP MTC network with respect to massiveaccess, reliable transmission and energy management are summarized in [38]. Forthe smart metering example mentioned above, the works in [39,40] provide a detailedassessment of the uplink traffic generated by such applications and develop a compre-hensive analytical model (inclusive of random access, control and data channels withretransmissions) to determine the meter outages. A downlink group-based pagingprocedure for smart meters is discussed in [41].In this thesis, we consider the physical layer aspects of low cost, low power, low10Chapter 1. Introductiondata rate MTC devices in LTE/LTE-A. The 3GPP has identified various categoriesof MTC devices [17,42].• Category-1 (CAT-1) - This category exists from the initial release of LTE(Release 8). A CAT-1 User Equipment (UE) can support date rates up to10 Mbps in the downlink and 5 Mbps in the uplink with a maximum bandwidthof 20 MHz.• Category-0 (CAT-0) - Release 12 of the LTE standard introduced CAT-0UEs, which support a maximum data rate of 1 Mbps in both the downlink andthe uplink, since the modulation is restricted to QPSK. This was the first stepby 3GPP towards improving the support for low data rate IoT devices. TheCAT-0 UE still supports bandwidths up to 20 MHz. However, the complexityof CAT-0 UEs is reduced to 50% of CAT-1 UEs.• Category-M1 (CAT-M1) - The first reduced-bandwidth device categoryintroduced by the 3GPP is the CAT-M1, introduced in Release 13 of theLTE/LTE-A standards. A CAT-M1 UE has a bandwidth of 1.4 MHz, which thelowest possible bandwidth supported by traditional LTE standards. A CAT-M1 UE has 75-80% reduction in complexity when compared to CAT-1 UEs andsupports data rates up to 1 Mbps.• Narrowband Internet of Things (NB-IoT) - Release 13 of the LTE/LTE-A standards also introduced the first narrowband device category in the formof NB-IoT. An NB-IoT UE occupies a bandwidth of 180 kHz and supports datarates up to 0.2 Mbps.In addition to the support for different categories of MTC UEs, MTC UE opera-tion with Coverage Enhancement (CE) is recognized as a “work item” in Release 1311Chapter 1. Introductionof the LTE standard, aiming to provide 12 dB to 20 dB of additional coverage [43,44].Moreover, the LTE/LTE-A standardization activities have identified that improve-ments to the current power saving mechanisms are necessary for efficient operationof MTC UE intending to support the IoT. In this thesis, we focus on suggestingenhancements to the Discontinuous Reception (DRX) mechanism, which is used forreduced power consumption in the downlink and the NB-IoT transmission mechanismin the uplink [45–48].1.1.3 Thesis Outline and Major ContributionsThe objective of our research is to provide mechanisms in the downlink and the uplinkfor facilitating the IoT using 3GPP LTE MTC. Our solutions are designed such thatthe changes required to the current LTE/LTE-A framework are kept minimal. In thefollowing, we provide an outline of the thesis, highlighting the major contributions.The rest of this chapter is organized as follows. In Section 1.2, we provide theessential background on the DRX and the NB-IoT mechanisms in LTE. This is fol-lowed by a review of the prior works on power saving modes in the downlink andenergy efficient transmission mechanisms in the uplink in Section 1.3. Chapter 2 andChapter 3 focus on energy efficient mechanisms for MTC UEs in the downlink, whileChapter 4 concentrates on such mechanisms in the uplink.The LTE/LTE-A standardization activities have recognized that the current DRXmechanism is not entirely efficient for MTC UEs, since it requires the UEs to periodi-cally check for the paging information, which is computationally intensive and powerconsuming. In Chapter 2, we propose a modified DRX mechanism incorporatingquick sleeping for energy efficient IoT using LTE/LTE-A [49–51]. Our contributionsin this regard involve the design of DRX with quick sleeping indication techniques12Chapter 1. Introductionfor two categories of low-mobility MTC UEs - a) normal coverage and b) extendedcoverage. For the former case, we develop the solutions by utilizing the existing re-sources on the physical downlink synchronization and the broadcast channels. Forthe latter case, we identify that in addition to the paging decode process, the UEresynchronization in the downlink also increases the power consumption. To alleviatethis problem, we develop solutions using dedicated resources on the physical downlinkdata channel, ensuring that the resource overhead is minimal.In Chapter 3, we explore the impact of timing reacquisition on the MTC UE en-ergy consumption in greater detail. We demonstrate that legacy methods for resyn-chronization in Orthogonal Frequency Division Multiplexing (OFDM) systems usingCyclic Prefix (CP) autocorrelation and reference signal detection are not effective forthe low complexity MTC UEs requiring extended coverage. Our main achievementin this chapter is design of the enhanced Primary Synchronization Signal (ePSS); anew signal aiding fast resynchronization and improved energy efficiency of the UE inthe downlink [52,53]. In the process, we also designed a new DRX mechanism, whichuses the ePSS as a quick sleeping mechanism, enabling us to harness the benefits ofboth faster reacquisition and simplified paging decode for further improvement in theenergy efficiency of the MTC UEs.Chapter 4 considers the MTC mechanisms under low network coverage for energyefficient uplink transmission in the realm of NB-IoT. We illustrate that the reduc-tion in residual Carrier Frequency Offset (CFO) improves data decoding at the basestation, which results in a reduction in the number of data retransmissions, therebyreducing the energy consumption of the UEs. Our major contribution in this chapteris to develop a maximum likelihood based CFO estimation mechanism for NB-IoTuplink, which works robustly in low coverage [54, 55]. We also demonstrate that the13Chapter 1. Introductionperformance of the large transport block transmission scheme (a novel technique be-ing considered by 3GPP for improved uplink data rates) is further improved whenused along with our proposed CFO estimation technique.Finally, the conclusions and potential avenues for further research are presentedin Chapter 5.1.2 Background1.2.1 DRX and Paging Mechanism in LTE/LTE-AIn the DRX mode of operation, the UE follows a periodic cycle called the DRX cycleinvolving sleep intervals and wake-up intervals. The current LTE/LTE-A standardssupport different DRX cycle lengths with a maximum length of 2.56 s. Recently,extended DRX cycle lengths starting from 5.12 s to a maximum of 2621.44 s havebeen approved by the 3GPP [56]. In this section, we review the DRX and pagingdecode mechanisms in the current LTE/LTE-A standards. Current Paging Reception Mechanism for UE in DRXThe paging information in LTE/LTE-A consists of the control part and the datapart. The control information for an upcoming paging block is indicated to the UEby a Physical Downlink Control Channel (PDCCH) containing the Paging - RadioNetwork Temporary Identifier (P-RNTI). This is followed by the paging data on thePhysical Downlink Shared Channel (PDSCH). The paging data consists of a list ofthe System Architecture Evolution - Temporary Mobile Subscriber Identity (S-TMSI)or the International Mobile Subscriber Identity (IMSI) of the UEs being paged. If theUE successfully decodes a PDCCH with P-RNTI and the paging block on PDSCH14Chapter 1. Introductionand finds its S-TMSI or IMSI in the paging list, then it stays awake to decode animpending data transmission. Otherwise, it goes back to sleep.For decoding the PDCCH, the UE requires a Discrete Fourier Transform (DFT)whose size corresponds to the base station (called the evolved Node B (eNB) in LTE)bandwidth. The UE also incorporates a blind decoding scheme where it hypothesizesover 44 options of PDCCH locations [57, 58]. This renders the PDCCH decodingprocedure to be computationally intensive and power consuming. The 3GPP stan-dardization committee is in the process of defining a new control channel for thelow-cost, low-complexity devices so that the number of blind decodes required todetect the paging identifier on the control channel is reduced [56]. But the processof looking for a page in each wake-up cycle is still retained. DRX Modes Supported in LTE/LTE-AThe LTE/LTE-A standard supports two variants of DRX - a) Connected Mode DRXand b) Idle Mode DRX [57,59,60]. These modes are categorized based on the handlingof the Radio Resource Configuration (RRC) connection. In the Connected ModeDRX, the UE does not relinquish the RRC connection for the entire duration of theDRX cycle and in the Idle Mode DRX, the UE releases the RRC connection beforeit goes back to sleep.In Connected Mode DRX, the ON time of the UE is 100 ms typically. LTEsubframes are 1 ms long and PDCCH is sent every subframe (see Figure 1.2). Hence,the UE monitors PDCCH for 100 subframes and a valid PDCCH is received on onlyone subframe [57,59]. Such a mechanism is beneficial to the UEs that receive a validPDCCH early during their ON time. They can process the paging information, goback to sleep quicker and save power.In the Idle Mode DRX, each UE checks for PDCCH periodically, albeit only in15Chapter 1. IntroductionSF 0 SF 1 SF 5 SF 91msLTE Radio Frame -10msSF 0 SF 5PDSCHPDCCHCRSSSS (SF 0)SSS (SF 5)PSSSlot 0 Slot 10.5ms 0.5msFrequencyPBCH(a) Physical channelsPBCH PSS and SSSPDCCHSF 0 SF 4 SF 5 SF 9. . . SF 0PDSCH Paging1ms1.4MHzSlot 1Slot 0(b) Physical channels with paging blocksFigure 1.2: LTE radio frame structure showing the different physical channels andpaging pre-assigned subframe per DRX cycle called the Paging Occasion (PO) [60]. ThePO subframe number is determined based on the UE Identifier (UEID) [60] and it canbe either subframe 0, 4, 5 or 9 as shown in Figure 1.2. This can provide significantpower savings compared to the Connected Mode DRX. Ideally, it can reduce the ONtime of the UE to just 1 ms if the SNR is good and if the timing synchronization ofthe UE is so accurate that it can wake up exactly at the PO. However in practice,the Voltage Controlled Oscillator (VCO) used for the UE clock will possess a drift16Chapter 1. Introductionthat can affect the symbol timing accuracy and SNR may not always be favourable.If the subframe timing is not accurate, the UE will have to re-acquire it using thePrimary Synchronization Signal (PSS) and the Secondary Synchronization Signal(SSS). Similarly, if the frame timing is lost, the UE will have to reacquire framesynchronization by decoding the System Frame Number (SFN) transmitted on thePhysical Broadcast Channel (PBCH) [58].Moreover, the computational complexity is substantial, since the UE is still re-quired to decode the PDCCH. The PDCCH is transmitted on 2, 3 or 4 OFDM symbolsdepending on the system bandwidth [57]. The UE has to search the different possiblelocations over the complete bandwidth to find its PDCCH. This requires a full-scaleDFT and blind decoding over 44 possible options of PDCCH locations [57,58] whichconsumes significant amount of UE processing power. Although the new controlchannel definition for low complexity devices can reduce computational complexityfor decoding of the paging control information [43], they would still require multi-ple repetitions of the control information in order to successfully decode the pagingidentifier owing to the low operating SNR. Furthermore, this decoding has to per-formed on each wake-up occasion of the DRX cycle, regardless of whether the devicehas a valid upcoming page or not, thereby increasing the energy consumption of thedevices.Figure 1.3 shows the different states traversed by a legacy UE implementing thepresent Idle Mode DRX mechanism. The UE is initially in a “Deep Sleep” state whereonly the UE clock is active and all other processing units including the radio are OFF.At the wake up time instant, the UE wakes up from DRX and resynchronizes withthe eNB by detecting PSS/SSS and PBCH (if the UE timing has drifted more thanhalf a subframe). After synchronization, the UE transitions to a “Light Sleep” state17Chapter 1. IntroductionRe-Sync – PSS/SSS DetectionDeep SleepPaging DecodeLight SleepWait for PO PO Sub-frameNo Page ReceivedPage ReceivedDRX Wake Up Connected StateFigure 1.3: Current DRX model in LTE/LTE-A.where only the clock, radio and channel estimation blocks are ON and the Tx/Rxprocessing blocks are OFF. The duration of light sleep depends on the PO. If the POcorresponds to the PSS/SSS subframe (i.e. on subframe 0 or 5), the UE proceedsto the “Paging Decode” state immediately (see Figure 1.2). But if the PO is onsubframe 4 or 9, the UE waits for the PO by moving to the “Light Sleep” state and ittransitions to “Paging Decode” state on the PO subframe. In the “Paging Decode”state, the UE decodes the PDCCH and the PDSCH (if the PDCCH contains theP-RNTI) for the paging information and moves to the “Connected State” if there isa valid page. Otherwise, it goes back to the “Deep Sleep” state.Modified DRX mechanisms for improved energy efficiency are described in Chap-ter 2 and Chapter 3 of this thesis.18Chapter 1. IntroductionSubframe =1msSlot 0 Slot 10.5ms 0.5msSub-carriers1 Slot =  7 symbolsPRB = 12 Sub-carriers x 7 Symbols (Normal CP)Sub-carriersData Symbols DMRS SymbolsFigure 1.4: Uplink subframe and PRB in LTE/LTE-A.1.2.2 Narrow-Band Internet of Things (NB-IoT) inLTE/LTE-AWith a large number of MTC devices requiring the cellular network for operationand a substantial portion of these devices being deployed in areas with bad networkcoverage, the 3GPP is in the process of standardizing the procedures for optimaloperation of such UEs. The research activities in this domain have been categorizedas NB-IoT and various mechanisms in downlink and uplink are being analyzed toaddress the requirements of MTC UEs [45–48]. In this section, we describe the NB-IoT transmission mechanism being standardized in the 3GPP LTE/LTE-A uplinkand analyze the energy efficiency of the MTC UEs using this mechanism.The basic unit of resource allocation in LTE/LTE-A is a Physical Resource Block(PRB). Considering a system with normal CP, one PRB consists of 12 subcarriers× 7 symbols (see Figure 1.4). Therefore, a PRB pair spans 12 subcarriers × 14symbols = 12 subcarriers × 1 subframe [57, 58, 61, 62]. Since the subcarrier spacingin LTE/LTE-A is 15 kHz, a PRB pair occupies a bandwidth of 15 × 12 = 180 kHz.The current LTE/LTE-A standards support UE transmission over multiple PRBs.However, considering that the MTC UEs are low data rate and low power devices,19Chapter 1. Introductionthe 3GPP has proposed the use of a single PRB pair transmission scheme for NB-IoTMTC UEs. Also, the modulation supported by these devices is restricted to QPSK.LTE/LTE-A has allowed the use of sub-PRB transmission, where the UE usesless than 12 subcarriers for its transmission. For example, the UE can adopt a single-tone transmission scheme, where the UE uses 1 subcarrier × 1 subframe transmissionand occupies a bandwidth of only 15 kHz [46–48]. Although using a single subcarrierreduces the data rate, it can still be effective for MTC UEs that are delay tolerant andonly require occasional small bursts of data to be transmitted. Furthermore, whenMTC UEs are in low coverage, the operating SNR at the eNB is very low (around-15 dB). Consequently, in the uplink, the UE has to transmit multiple repetitionsof the data block to be successfully decoded by the eNB, thereby increasing the ONtime and the energy consumption of the UE. Therefore, identifying signal processingtechniques that can reduce the number of repetitions is necessary.In LTE/LTE-A, a data block at the physical layer is called the transport block.The transport block is encoded using a Convolution Turbo Code (CTC) before trans-mission and 4 Redundancy Version (RV)s are generated [58,59]. One RV of the trans-port block is transmitted in one subframe. Each RV includes a 24-bit header fromthe upper layers [58,59]. Therefore, the effective data rate is given byReff =(TBS− 24)(NSF × tSF) (1.1)where TBS is transport block size, NSF is the number of subframes required by theeNB to successfully decode the data and tSF = 1 ms, is the duration of a subframein LTE/LTE-A. In the case of low data rate MTC UEs, the transmission consists ofa burst of data packets followed by a long idle duration. When the effective datarate increases, the UE can complete its data transmission quickly and switch to the20Chapter 1. Introductionidle mode sooner, thereby saving power. For a given Transport Block Size (TBS),the effective data rate increases if the number of subframes required for successfuldecoding decreases. This depends on the SNR, the underlying channel and the offsetin UE’s timing/frequency estimation at the eNB.The UE timing/frequency offset is derived from the detection of the random accesssignal transmitted by the UE when it first requests network access and/or the periodicDemodulation Reference Signals (DMRS) transmitted by the UE in every subframe[58]. Both the random access and DMRS signals are Zadoff-Chu (ZC) sequences,which possess good detection properties (good autocorrelation, low cross-correlation)[58, 59]. However these estimates are not perfect and there will be some residualtiming and frequency offset in the system. The effective data rate and hence theenergy efficiency of the UE can be improved if these residual offsets are reduced to anegligible level.The residual timing offset can be estimated with a sufficient degree of accuracyusing the CP [63, 64]. The eNB ensures that all UE transmissions are time synchro-nized using the timing advance indication mechanism after the initial random accessrequest procedure in LTE/LTE-A and minor deviations in the received frame timingare tracked using CP autocorrelation [58]. However, the residual CFO of each UEmight be different and tracking each UE’s CFO using CP autocorrelation is complex.Therefore, the eNB needs a separate mechanism to compensate for this CFO andimprove the energy efficiency of the UE.In Chapter 4 of this thesis, a robust ML based CFO estimation algorithm, whichoutperforms the conventional CP autocorrelation method is demonstrated to improvethe energy efficiency of the MTC UEs for the NB-IoT uplink.21Chapter 1. Introduction1.3 Literature ReviewWith the necessary background information established from the previous section, inthe following, we review the prior works elated to the energy efficient mechanisms inthe downlink and the uplink across different wireless communication technologies.1.3.1 Prior Work on Power Saving Mode and DRXAnalysisThe study of the performance of the power-saving mode of operation in wirelesscommunication systems has gathered the interest of researchers in various technicalcommunities.The power-saving mode supported by the IEEE 802.16 standard for WirelessInteroperability for Microwave Access (WiMax) has been analyzed in [65–71]. Initialwork on analyzing the sleep mode in the IEEE 802.16 e standard was carried outin [65], where the authors determine the packet dropping probability and the meanwaiting times of packets in the base-station buffer for Poisson packet arrival processand a general distribution for service time. An analytical model to determine thesleep mode energy savings was developed in [66], assuming only incoming messagesand a Poisson distribution for the frame arrival rate. In [67], the authors extendthe work in [66] by considering the effects of both the incoming frames and outgoingframes on the sleep time. These queueing theory based models were adopted for IEEE802.16 m standard in [68] (for uncorrelated traffic) and [69] (for correlated traffic)by incorporating the duration of the awake state and a close down timer indicatingthe end of the awake state. In [70], the authors proposed an efficient power-savingmechanism involving periodic traffic indications, which is also resource-efficient forIEEE 802.16 e/m. The analysis of the power-saving mode in terms of the trade-off22Chapter 1. Introductionbetween the power consumption and the packet transmission delay for heterogeneoustraffic (real time and non-real time) was carried out in [71].The term “DRX” for power-saving was first introduced in the Universal MobileTelecommunications System (UMTS) standard. In [72], the authors investigated theUMTS DRX mechanism for reduced power consumption using two parameters - theinactivity timer threshold and the length of the DRX cycle. The inactivity timer is atimer initiated by the Radio Network Controller (RNC), when there is no packet tosend to the mobile device. The mobile device can go to “sleep” only after the inactivetimer has expired. If a packet arrives at the RNC before the inactivity timer expires,then the link moves to“busy” state. A variant of the M/G/1 queueing mode withvacations was presented in [72] to analyze the UMTS DRX performance in terms ofthe expected queue length, the expected packet waiting time, and the power savingfactor. This work was extended in [73], where an adaptive algorithm called dynamicDRX was devised to dynamically adjust the inactivity timer threshold and the DRXcycle length values to enhance the performance of the UMTS DRX. An adaptiveDRX mechanism for UMTS was developed in [74]. Here, the DRX period for eachUE was individually and adaptively controlled by the UMTS base-station using theextended paging indicator, which in-turn was configured based on the current trafficsituation for each UE. The aforementioned works on the UMTS DRX consideredPoisson traffic. In [75], the authors proposed a novel semi-Markov process to modelthe UMTS DRX with bursty packet data traffic. The study provided the selectionguidelines for the inactivity timer and DRX cycle values under various packet trafficpatterns. The study of the power consumption and the mean packet waiting timeof a dual-mode mobile device (a device that supports both UMTS and WLAN radiotechnologies) is conducted in [76].23Chapter 1. Introduction1.3.1.1 Prior Work on LTE DRXWith the advent of LTE as a leading wireless communication standard, the study ofthe DRX mechanism in LTE has gained a lot of interest over the past few years. TheLTE DRX operation and directions towards improving the DRX parameter selectionwas provided in [77]. In [78], the DRX mechanism is modeled for bursty packet datatraffic using a semi-Markov process. The power saving and wake-up delay perfor-mance of the LTE DRX mechanism are evaluated and compared against the UMTSDRX mechanism. It is shown that the LTE DRX always achieves better power sav-ing performance at the cost of longer wake-up delay. The LTE/LTE-A standardssupport two types of DRX based on the cycle length - short DRX and long DRX,whose performance in terms of user throughput, power consumption, and networkperformance is evaluated in [79].In [78] as well as the references analyzing the UMTS DRX (see [72–75]), theprocedure followed to obtain the power-saving factor is quite complex, requiring thedifferentiation of Laplace transform equations. This complexity is avoided in [80],where the authors provide a numerical analysis of the DRX by dividing the operationinto several independent parts and combining the result obtained from each part.The next line of interest in LTE DRX analysis is in determining the best setof DRX parameters, such as the inactivity timer duration and cycle length, for theoptimal performance of a particular application. In [81], a web-browsing session asthe underlying application and different algorithms for optimizing the balance amonguser throughput and power saving is obtained. The optimization of LTE DRX formobile Internet applications over is considered in [82] and an algorithm is proposedto efficiently select DRX parameters to ensure a balanced trade-off between twoconflicting performance parameters - application delay and UE power savings. In [83],24Chapter 1. Introductionan analytical model for LTE DRX based on a semi-Markov process is developed toprovide new, simple and exact formula relevant for the power saving efficiency. Thepaper also investigated the LTE DRX performance with bursty packet data trafficand the effects of different DRX parameters on the power saving and wake-up delay.The evaluation of the influences of the Transmit Time Interval (TTI) sizes and theeffects of LTE DRX Light and Deep Sleep mode on power consumption for voiceand web traffic has been carried out in [84]. The work in [85] investigated the use ofadjustable and non-adjustable DRX cycle frame duration in LTE for reduced powerconsumption based on a semi-Markov model for bursty traffic. A model to analyze thelatency incurred by the DRX mechanism for active and background mobile traffic isdeveloped in [86]. The paper also proposed a mechanism to switch DRX configurationbased on traffic running at UE. Recently, an in-depth analysis of the average delayand power consumption of the DRX mode adopting recursive deduction and Markovmodels is provided in [87]. The paper presented the analysis using mixed DRX shortand long cycles and Poisson packet arrival.From the MTC perspective, in [88], a semi-Markov chain model is developed toanalyze the DRX mechanism for MTC applications, which can be used to estimatethe choice of DRX parameters. The LTE DRX performance for MTC was evaluatedin [89], where the authors incorporate a model with different parameter settings cor-responding to potential future M2M devices. The results indicate that extending themaximum DRX cycle length would lead to significant improvement in the energy ef-ficiency of M2M devices. The work in [90] also suggested the benefits of an extendedDRX cycle for MTC devices and analyzed the impact of introducing the extendedDRX feature on the operation of the networks, thereby providing a guideline to facil-itate extended DRX. A new approach of augmenting the extended DRX mechanisms25Chapter 1. Introductionwith the single-packet-active state embedded in the sleep cycle was presented in [91].This technique is shown to not only improve the power saving obtained from theDRX but also reduce the wake-up delays compared to the standard DRX methodbecause it is sensitive to the underlying traffic pattern and able to shift between theactive and sleep states rapidly.However, all these works address the DRX operation for UEs in normal coverageand assume perfect timing synchronization. For the low-complexity MTC UEs withcoverage enhancement, the authors in [92] demonstrate a new mechanism where theUE does not check for the page periodically and turns ON its radio only for datatransmission, thereby reducing the energy consumption of the UE. This mechanismis only applicable to transmit driven UEs and cannot be used by UEs which requiretimely information from the eNB to operate successfully.In Chapter 2 and Chapter 3 of this thesis, we elaborate on our modified DRXmechanism, which improves the downlink energy efficiency of MTC UEs in nor-mal coverage and low coverage. These mechanisms require a minimum bandwidthof 1.4 MHz and are therefore applicable to the UE categories of CAT-M1, CAT-0and above. However, the ideas can be easily modified to suit devices with smallerbandwidths and will serve as a medium to design novel power saving and pagingmechanisms in the downlink for the upcoming 5G New Radio (NR) standard.1.3.2 Prior Work on Energy Efficient M2M UplinkMechanismsThe uplink is considered to be the most critical channel for M2M communication,especially with the increasing number of IoT M2M devices. The design and devel-opment of energy efficient mechanisms for M2M uplink has been studied in various26Chapter 1. Introductionprevious works.An optimized signal flow in the uplink for M2M devices, radio access networksand core networks was proposed in [93]. The paper suggested the use of simplifieduplink mechanisms at the M2M device, such as reduced Channel State Information(CSI) feedback for 3GPP MTC UEs to conserve power. In [94], an access controlalgorithm using K-means based grouping and coordinator selection is proposed toreduce the uplink energy consumption and alleviate the loading at the base-stationoccurring from the massive access requests from the M2M devices.The effect of the large number of M2M devices on the random access channelhas been analyzed in [95–98]. In [95], the problem of radio access network overloadis addressed using a prioritized random access scheme. The proposed scheme alsoensures the QoS constraints of the different classes of the LTE MTC devices bypre-allocating the Random Access Channel (RACH) resources for the different MTCclasses, defining the class-dependent backoff procedures and by using dynamic accessbarring (see [96]) to prevent a large number of simultaneous RACH attempts. Anevaluation methodology fully compatible with the 3GPP test cases is proposed in [97],which incorporates a thorough analysis of RACH performance in overloaded MTCscenarios. The work presents an analytical model for RACH including the energyconsumption aspect along with the conventional performance metrics, such as theaccess delay and the collision probability. In [98], a reinforcement learning-based eNBselection algorithm is proposed that allows the MTC devices to transmit packets ina self-organized manner to the chosen eNB.The energy efficiency aspect with respect to MTC data transmission has been ad-dressed in [99–102]. Joint massive access control and resource allocation schemes areproposed in [99], which perform machine node grouping, coordinator selection and27Chapter 1. Introductioncoordinator resource allocation. The proposed schemes consider a two-hop transmis-sion protocol and determine the number of groups required to minimize total energyconsumption under both flat fading and frequency-selective fading channels. Forsmall data transmission, a novel contention-based LTE transmission mechanism isproposed in [100]. The performance results demonstrate reduced network resourceconsumption, shorter mean data delay and improved device energy efficiency. In [101],it is demonstrated that the energy-efficiency of the transmissions of small data blocksstrongly depends on the transmission power and the adaptive modulation and codingprocedure used. The authors propose a scheme to determine the optimal Modulationand Coding Scheme (MCS) for small data transmission utilizing the LTE uplink powercontrol mechanism. The work in [102] focuses on developing a systematic frameworkto study the power and energy optimal system design for M2M. The optimal trans-mit power, energy per bit, and the maximum load supported by the base station arederived for a variety of coordinated and uncoordinated transmission strategies. Itis demonstrated that Frequency Division Multiple access (FDMA), including equalbandwidth allocation, is sum-power optimal for the low spectral efficiency regimeand the performance of uncoordinated Code Division Multiple Access (CDMA) iscomparable to FDMA when the base station is lightly loaded.The aforementioned works have only been evaluated for UEs under normal cov-erage. For UEs in low coverage, several TTI bundling enhancement schemes areproposed in [103] based on the parameterization of the bundle size, round trip timeand the maximum number of HARQ retransmissions. These schemes are evaluatedusing link-level simulations and shown to improve the LTE uplink coverage. In [104],the TTI bundling enhancement, frequency hopping, increased number of base-stationreceiver antennas and power boosted uplink reference signals have been considered as28Chapter 1. Introductionpotential candidates for coverage enhancement in the LTE uplink. In [105], a flexibleTTI bundling scheme with CDMA support us demonstrated to improve the coverageof LTE MTC devices.Chapter 4 of this thesis focuses on NB-IoT based uplink transmission mechanismscombined with robust CFO estimation for enhancing the energy efficiency of theMTC UEs in the uplink. Since these mechanisms are designed for the lowest possiblebandwidth supported by the LTE MTC standard, it is applicable to all the UEcategories (NB-IoT, CAT-M1 and above) and the design principles can be adoptedfor the forthcoming 5G NR standard.29Chapter 2Discontinuous Reception (DRX)with Quick Sleeping2.1 IntroductionThe current LTE/LTE-A standards use the DRX mechanism to reduce the powerconsumption of the UE during which it follows a sleep and wake-up cycle (as explainedin Section 1.2.1). The UE wakes up periodically to check for the paging informationfrom the eNB. Though the DRX procedure results in significant power reduction, theamount of power spent by the UE during the wake-up time or the ON time is stillconsiderable, since the decoding of paging information is computationally intensive.Also, the UE may need to reacquire timing synchronization due to the drift in theUE clock, which further increases the ON time. Moreover, in the IoT scenario, due tothe presence of a huge number of devices, the probability of the UE receiving a pageduring each ON period is substantially low. However, the UE still looks for paginginformation every time it wakes up and expends a significant amount of power.In this chapter, we propose the Quick Sleeping Indication (QSI) mechanism toindicate to the UE whether it can sleep early, since there is no valid incoming page.When our QSI mechanism is used, the UE would first decode the QSI message andif it indicates “sleep”, the UE goes back to sleep immediately. If the QSI indicates“stay-awake”, the UE remains ON for decoding the subsequent paging block.30Chapter 2. Discontinuous Reception (DRX) with Quick SleepingIt would be helpful to the UE if the QSI mechanism is simple and structured suchthat the UE can decode it with low complexity and reduced power consumption. Tothis end, we provide the QSI design mechanisms for two categories of MTC UEs - a)without CE and b) with CE. Our QSI mechanisms are not only simple to implement,but also require minimal changes to the present 3GPP LTE/LTE-A standardizationframework. We show that we can obtain a substantial improvement in energy effi-ciency and a significant reduction in computational complexity using our novel QSImechanisms for MTC UEs with and without CE.The rest of the chapter is outlined as follows. In Section 2.2, we introduce the QSImechanism and explain our modified DRX with QSI. In Section 2.3, we discuss theQSI mechanisms for MTC UEs without CE. In Section 2.4, we consider the case ofMTC UEs with CE and demonstrate our QSI mechanisms for this case. We follow upwith the energy efficiency and computational complexity analysis in Section 2.5 andthe simulation results in Section 2.6. The conclusions are presented in Section Details of the QSI MechanismIn the IoT scenario, the eNB would have to communicate the paging message to alarge set of MTC UEs. The paging message is transmitted in the downlink through atransport channel called the paging channel, which is mapped to the PDSCH physicalchannel. The paging channel can accommodate a maximum of 16 UE identities[59]. Thus, the eNB would have to schedule paging information multiple times,which would lead to an increase in the time required by an UE to receive the paginginformation. The MTC UEs might be able to handle the delay in receiving paginginformation, since they are delay tolerant. However, the UEs listen to multiple pagingoccasions before they receive a valid paging message resulting in increased power31Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingconsumption.If a QSI mechanism is introduced, then it gives an indication of whether theUE can go back to sleep early since it does not have an impending paging messageor stay awake to decode the PDCCH. Each UE decodes the QSI and processes thePDCCH only if the QSI inhibits it from sleeping early. In this section, we describeour QSI mechanism and propose DRX with quick sleeping for energy efficient IoTusing LTE/LTE-A.2.2.1 QSI Working MechanismOne method that the eNB can adopt to address a large number of UEs is to dividethe UEs into multiple groups and allocate resources to one or more groups at eachscheduling interval [43]. We propose to group the UEs using some unique UE identi-fier. Let us consider that the UEs have to be divided into Ngrp groups. A unique UEidentifier can be used to determine the UE Group Indicator (UEGI) by the relationUEGI = < Unique UE Identifier > mod Ngrp. The following identifiers can beused to determine the UE Group:1. IMSI - The IMSI is a unique identifier assigned to every UE during the equip-ment manufacturing phase.2. UEID - It is given by UEID = IMSI mod 1024 and it is being used todetermine the paging occasion in Idle Mode DRX [60].3. Connected Mode - Radio Network Temporary Identifier (C-RNTI) - The C-RNTI is a unique number assigned to each UE when it establishes a connectionwith the eNB.The device grouping techniques are out of the scope of our work. Therefore, we do32Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingnot elaborate on the device grouping strategies in MTC. We assume that the UEsare divided into groups using a known device grouping strategy and introduce theQSI mechanism. The proposed QSI works as follows:1. The UEs are divided into Ngrp groups.2. QSI is sent n subframes before the paging occasion where n is determined bythe QoS constraints of the MTC UE group. It indicates one or more UE groupswhich will not be paged in the paging occasion.3. The UE decodes QSI and if it indicates that its group will not be paged, theUE goes back to sleep. Otherwise, the UE prepares to decode PDCCH on thepaging occasion.After we divide the UEs into multiple groups, we assign M bits for the QSImessage which would put one or more UE groups to sleep when there is no impendingpage for the group. The network has the flexibility to configure the manner in whichthe QSI addresses the UE groups. One configuration that the network can use is abit-map addressing mode where one bit is assigned exclusively to one UE group andthe network can address up to M groups. For example, with M = 4, we can address4 UE groups and the QSI message “1010” indicates that the UE groups 1 and 3 canbe put to sleep.It should also be noted that the UE in Idle Mode DRX cannot directly wake up atthe paging occasion to decode PDCCH since it has to acquire timing synchronizationand update the SFN. The frame timing and synchronization is achieved by detectingthe PSS and the SSS [61]. The SFN is obtained by processing the PBCH and decodingthe Master Information Block (MIB) [61].33Chapter 2. Discontinuous Reception (DRX) with Quick SleepingRe-Sync – PSS/SSS DetectionDeep SleepPaging DecodeQSI DetectionPO Sub-frameNo Page ReceivedDRX Wake Up Light  SleepQSI indicates “Sleep”Page ReceivedConnected StateQSI indicates “Stay-Awake”(a) Proposed DRX model with QSI (without CE)QSI and Timing DetectionDeep SleepPaging DecodeNo Page ReceivedDRX Wake Up QSI indicates “Sleep”Page ReceivedConnected StateQSI indicates “Stay-Awake”Light SleepPO Sub-frame(b) Proposed DRX model with QSI (with CE)Figure 2.1: Proposed DRX models.34Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping2.2.2 DRX and Paging with QSIFigure 2.1a demonstrates our first model for DRX and paging with QSI, which weuse for MTC UEs without CE. In this model, the UE begins its paging detectionoperation similar to the legacy UE, by transitioning from the “Deep Sleep” stateto the “Re-sync State” to acquire the symbol boundary when it wakes up from theDRX cycle. However, after the timing acquisition, the UE transitions to the “QSIDetection” state, where it detects the QSI signal. If the QSI conveys “sleep” sincethere is no valid upcoming page, then the UE immediately transitions into the “DeepSleep” state. However, if the QSI signals indicates “stay-awake” or if the QSI is notdetected successfully, the UE resumes the legacy operation for decoding the paginginformation and transitions to the “Light Sleep” state.The MTC UEs can be deployed in places like interiors of buildings and basementswhere the network coverage is low. When the SNR is low, multiple PSS/SSS copieswill have to be combined for successful detection and timing synchronization. ThePSS/SSS is transmitted every 5 ms and the UE has to stay ON longer if it requiresmultiple copies of PSS/SSS. Also, the PDCCH and PDSCH decoding might requiremultiple repetitions to be decoded successfully which would further increase the ONtime and the computational complexity for paging decode. Therefore, it is beneficialto have a QSI signal which can also be used for subframe synchronization for MTCUEs with CE. In this case, the different combinations of M QSI bits are mappedto different sequences, thereby resulting in 2M QSI sequences. The sequence corre-sponding to the QSI message is transmitted periodically. The UE receiver detectsthe transmitted QSI sequence by hypothesizing over the set of 2M QSI sequences anddecodes the QSI message as well as the timing information.Figure 2.1b depicts our second model for DRX and paging with QSI for CE. The35Chapter 2. Discontinuous Reception (DRX) with Quick SleepingUE begins its paging detection operation by transitioning from the “Deep Sleep”state to the “QSI and Timing Detection” state where it jointly obtains the timingand sleeping indication. If the QSI conveys “sleep”, the UE moves back to the “DeepSleep” state immediately and if the QSI indicates “stay-awake”, the UE transitions tothe “Light Sleep” state and decodes the page by moving to the “Paging Decode” stateon the PO. If the QSI is not detected, then the UE follows legacy DRX operation inorder to decode the paging information.In the following, we present different implementations for the proposed DRX andpaging with QSI for MTC UEs without CE and with CE. Our QSI solutions aredesigned such that they are compatible with the LTE/LTE-A framework and help inenergy efficient operation of MTC UEs for IoT.2.3 Quick Sleeping Solutions for MTC UEsWithout CEIn this section, we discuss the QSI mechanisms for MTC UEs without CE. This isapplicable to the IoT scenario of pet tracking or weather sensing in which the UEshave low-mobility and are located in regions where the network coverage is good. Inthis case, we design the QSI such that it reuses the resources that are already beingallocated by the eNB. We choose to transmit the QSI on those physical channelswhose locations on the subframe grid do not change so that the UE is aware of thelocation of the QSI. The physical channels for synchronization and broadcast complywith our requirement.The key feature of PSS, SSS and PBCH is that they always occupy a constantbandwidth of 1.4 MHz regardless of the system bandwidth. The PSS and SSS are36Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingtransmitted every 5 ms on subframe 0 and subframe 5 (see Figure 1.2) [57]. Theyoccupy one symbol and the UE can use a correlation decoder to detect these syn-chronization signals and adjust the frame timing [58]. The PBCH occupies 4 symbolson subframe 0 (see Figure 1.2) and is transmitted every 10 ms and has 4 repetitions.Therefore, a new PBCH is transmitted every 40 ms. The construction of PBCH blockis such that the UE can decode each 10 ms transmission independently or combinemultiple repetitions for decoding PBCH [61]. Due to smaller bandwidth, the UE canuse a smaller Fast Fourier Transform (FFT) size to decode the PBCH. Therefore,an early sleeping indication or QSI during the synchronization or the PBCH detec-tion phase would help the UE to determine if it has to go back to sleep or proceedto decode PDCCH decoding. Hence, we develop simple and efficient techniques totransmit QSI on PBCH and PSS/SSS.2.3.1 Quick Sleeping Solutions Using the PBCHFirst, we present a brief overview of the PBCH transmission in the current 3GPP LTEstandard. Then, we demonstrate four methods for incorporating the QSI mechanisminto the PBCH. PBCH in 3GPP LTEPBCH carries the MIB, which is 24 bits of information. As depicted in Figure 2.2, itindicates the eNB bandwidth, the Physical Hybrid ARQ Indicator Channel (PHICH)duration, the PHICH resolution and the SFN [12, 62]. The SFN in LTE has 10bits. Since the MIB has only 8 bits to represent SFN, this field it indicates 8 mostsignificant bits of SFN. From the PBCH processing in Figure 2.2, it can be seen thata 16 bit Cyclic Redundancy Checksum (CRC) is attached to the 24 information bits37Chapter 2. Discontinuous Reception (DRX) with Quick SleepingSystem Bandwidth (3 bits)PHICH Duration (1 bit)PHICH Resolution (2 bits)SFN  (8 bits)Reserved (10 bits)24-bit Master Information Block (MIB)24-bit MIBAdd 16 bit CRC1/3 rate Convolution Coding and Rate MatchingScrambling using Cell IdQPSK ModulationLayer Mapping and PrecodingResource MappingPBCH ConstructionRadio Frame (10ms)PBCH containing 240 QPSK symbols occupying 4 OFDM symbols on Sub-frame 0. 1 Sub-fame = 1ms 40 bits1920 bits1920 bits960 symbolsEncoding and ModulationFigure 2.2: MIB and PBCH construction.transforming MIB into a 40 bit block. This is followed by rate 1/3 Convolution Coding(CC) and rate matching so that the encoded MIB is 1920 bits. The modulation usedfor PBCH is QPSK and these modulated symbols have to be transmitted in 40 ms.This corresponds to 960 QPSK symbols every 40 ms which implies that there are 24038Chapter 2. Discontinuous Reception (DRX) with Quick SleepingQPSK symbols in every 10 ms copy of PBCH. QSI Using Reserved Bits in MIBOur first solution is to use the reserved bits in MIB for QSI. The MIB transmittedon PBCH contains 10 reserved bits out of the 24 information bits. Considering thatthe UEs are divided into groups, these bits can be used to indicate the UEGI. UsingM reserved bits, 2M UE groups can be addressed. For example, by using 4 out of10 reserved bits, we can address 16 UE groups. Alternatively, M reserved bits canbe used to indicate multiple UE groups. For example, QSI bit 0 would indicate UEgroups 1 to 4, QSI bit 1 would indicate UE groups 5 to 8 and so on. Using reservedbits in MIB ensures that QSI is decoded along with PBCH without affecting theperformance of PBCH decoding. Spreading QSI Using Orthogonal SequencesOur second solution is to transmit the QSI over PBCH using sequences that areorthogonal to PBCH. That is, the M bits of QSI are mapped to 2M sequences whichare orthogonal to the PBCH transmission. The orthogonal sequence corresponding tothe QSI message is sent along with the PBCH at a very low power to ensure that theloss in the power of PBCH is small. This method is illustrated in Figure 2.3, whereQSI modulated sequence denotes the orthogonal QSI sequence. Let Sref denote thepower of the PBCH signal without QSI, N denote the noise power and Sp denote thepower of the orthogonal QSI sequence. The SNR of the PBCH signal without QSI isSNRref =SrefNand the SNR of the QSI signal is SNRp =SpN. When the QSI signal isadded to PBCH, the power of the PBCH signal becomes Snew = (1−P )·Sref , where Pis the fraction of the PBCH power used for QSI. Therefore, SNRnew = (1−P ) ·SNRrefand the loss in PBCH detection performance is (1 − P ). For example, if P = 0.05,39Chapter 2. Discontinuous Reception (DRX) with Quick SleepingLayer Mapping and Pre codingResource MappingPBCH modulated signalQSI modulated signalScale Magnitude by sqrt(1-P)Choose the current 10ms PBCH copyPBCH Encoding and Modulation960 QPSK Symbols240 QPSK SymbolsScale Magnitude by sqrt(P)QSI + PBCH TransmissionFigure 2.3: PBCH with QSI transmissionthen SNRnew = 0.95 · SNRref and the loss in PBCH detection performance in dB is10 log10(0.95) = 0.22 dB, which is small. This loss occurs since we formulate our QSImechanism such that the total power available for PBCH transmission is unchanged.If the eNB can afford additional power for QSI transmission, there will be no loss inPBCH performance.The QSI sequences should be orthogonal to the different possible PBCH trans-missions. LTE supports 6 different eNB bandwidths (1.4 MHz, 3 MHz, 5 MHz,10 MHz, 15 MHz, 20 MHz), 2 PHICH durations (normal, extended), 4 PHICH res-olutions(16, 12, 1, 2)and an 8 bit field to indicate SFN [58]. This gives rise to6 × 2 × 4 × 28 = 12, 288 combinations. Additionally, there are 3 antenna config-urations (NTX = 1, 2 or 4) resulting in 12, 288 × 3 = 36, 864 combinations. Since40Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingthe PBCH symbols are scrambled by the Cell ID, there are 36,864 combinationsper Cell ID. Therefore, for a given Cell ID and a 10 ms copy of PBCH, we have a36, 284 × 240 complex matrix representing all the possible combinations of PBCH.We generated this matrix in MATLAB and observed that the rank of this matrix ispretty low (approximately 100). This suggests that sequences orthogonal to PBCHcombinations per Cell ID can be derived from the null space of this matrix. There are504 Cell Ids. Therefore, the total number of QSI sequences is 504 × 2M , each beinga complex vector of length 240. Also, there are 4 different 10 ms copies of PBCH.If each complex number is represented using 32 bits, this scheme would require atable of (4 × 504 × 2M × 240 × 4) bytes . For example, with 4 QSI bits, there are504×16 = 8064 QSI sequences and they would require a memory of 29.53 MB whichis large. In the following, we describe the QSI methods which eliminate the need fororthogonal sequences and hence does not require extra memory. Spreading QSI Using Repetition OnlyIn this method, the QSI bits are repeated at a very low power across the PBCHsymbols similar to the solution using orthogonal sequences. But in this case, the QSImodulated signal in Figure 2.3 is generated as shown in Figure 2.4a. The M bitsof QSI are spread over 480 bits of 10 ms copy of the PBCH. Hence the spreadingfactor SPF = 480M. The QSI bits are also modulated using QPSK. Due to the largespreading gain, the QSI signal can be detected after we subtract the detected PBCHsignal from the received signal. The SNR to decode QSI signal considering thatthe PBCH signal is subtracted from the received signal is P · SNRp · SPF . Forexample, with M = 4 and P = 0.05, SPF = 120 and the SNR to decode QSIwill be 6 × SNRp which is 10 log10(6) = 7.78 dB more the original QSI SNR. Inaddition to repetition, a randomizing sequence (RS) of length 240 symbols is used to41Chapter 2. Discontinuous Reception (DRX) with Quick SleepingM bit QSI messageQPSK ModulationRepetition (480/M)QSI modulated signalMultiply by the Phase Randomizing Sequence(a) Spreading QSI using repetition onlyM bit QSI messageFEC EncodingQPSK ModulationRepetition (480/Menc)QSI modulated signalMultiply by the Phase Randomizing SequenceMenc/2 QPSK SymbolsMenc bits(b) Spreading QSI using repetition and FECFigure 2.4: Illustration of QSI mechanisms on PBCH.ensure uniform distribution of phase of the QSI modulated signal. The transmittedconstellation is determined byT (n) =√1− P ejθ(n) +√P ejβ(n)ejφ(n) (2.1)where θ(n) is the phase of the nth PBCH modulated signal, φ(n) is the phase ofthe nth QSI modulated signal, β(n) is the phase of the nth complex number in therandomizing sequence and n = 0, 1, . . . 239. The Signal-to-Interference-plus-Noise42Chapter 2. Discontinuous Reception (DRX) with Quick SleepingRatio (SINR) is given bySINR =(1 − P ) · SrefP · Sp + N=(1 − P ) · SNRrefP · SNRp + 1 (2.2)Therefore, we expect that the loss in performance is the factor (1 − P )(P ·SNRp + 1) . Again,the loss occurs since we model the QSI mechanism such that the total power availablefor PBCH transmission is unchanged. Spreading QSI Using Repetition and Forward Error Correction(FEC)This solution is similar to the previous one, but the QSI bits are encoded using anerror correction coding scheme (see Figure 2.3 and Figure 2.4b). Here, the M QSIbits result in Menc encoded bits and the spreading factor is SPF =480Menc. The SNRto decode the QSI signal considering that the PBCH signal is subtracted from thereceived signal is reduced by a factor of MMenc, but the FEC makes up for the loss inSNR. Also, if the number of QSI sequences is small, a Maximum Likelihood (ML)decoding scheme can be used for FEC decoding. For example, with M = 4, P = 0.05and an (8,4) extended Hamming code for FEC, Menc = 8 and SPF = 60.Next, we discuss our QSI solutions using the synchronization channels (PSS/SSS).2.3.2 QSI on PSS/SSSPSS and SSS are the synchronization signals used in LTE/LTE-A transmitted on thecentre band. The PSS is a 63-length ZC sequence which is sent with a periodicityof 5 ms on the last symbol of the first slot in subframe 0 and subframe 5. The 32nd43Chapter 2. Discontinuous Reception (DRX) with Quick SleepingTable 2.1: QSI transmission methods using unused subcarriers on PSS and SSSMethod Cm FEC Used Ns Nr1 BPSK No 1 22 BPSK No 2 43 BPSK Yes 1 14 BPSK Yes 2 2carrier corresponds to the DC subcarrier and it is set to zero [58]. The SSS consists oftwo 31-length m-sequences on either side of the DC subcarrier. The SSS is also sentin subframe 0 and subframe 5 one symbol before the PSS. But the SSS on subframe0 is not the same as the one on subframe 5 and this helps the UE determine if it is onthe first half of the radio frame or the second half during acquisition [58]. The centreband spanning 1.4MHz consists of 72 subcarriers including the DC subcarrier. Thekey feature of both PSS and SSS transmissions is that they use only 62 out of the 72subcarriers. Thus, excluding the DC subcarrier, we still have 9 unused subcarriers.We therefore propose to transmit M QSI bits using 8 out of the 9 unused subcarriers.For this mechanism, we consider the case where M ≤ 4 and build a 4-bit QSImessage. When M ≤ 2, the bits can be repeated and when M = 3, a zero canbe appended as the most significant bit in order to obtain the 4-bit QSI message.Let Cm and Ns denote the modulation scheme and the number of synchronizationsymbols used for QSI transmission respectively. The number of repetitions requiredto accommodate the 4-bit message on the unused subcarriers is Nr =8·Ns4= 2Ns.Table 2.1 summarizes variants for the proposed QSI transmission on PSS/SSS usingdifferent repetition factors without and with FEC. For the latter, we suggest a (8,4)extended Hamming code, because we use 8 unused subcarriers per synchronizationsymbol to accommodate the 4-bit QSI message.We assume that the eNB has to use a part of the available power for transmittingthe QSI. When there is no QSI, the eNB transmission power is uniformly distributed44Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingover 62 subcarriers and in the presence of QSI, it is distributed uniformly over 70subcarriers. Therefore, the loss in PSS/SSS SNR when QSI is transmitted can becomputed as 10 log10(6270) = −0.53 dB, which is a small degradation from the originalvalue. If the eNB can afford additional power for QSI transmission, there will be nodegradation in PSS/SSS performance.2.4 Quick Sleeping Solution for MTC UEs WithCEIn the case of IoT, the UEs may be located in places like underground parking lots tosense vacant parking spots or in the interior of buildings such as hospitals to monitorthe status of the patients where the network coverage is very low. The solutionsdiscussed in Section 2.3 do not work effectively in this case, because the UE will needmultiple repetitions of PSS/SSS to determine the timing since the SNR is very low.And if the UE needs to re-acquire PBCH, it would need multiple copies of PBCH toaccurately determine the SFN. Similarly, decoding the paging on PDCCH, PDSCHand the QSI will also require multiple repetitions, which increases the ON time of theUE. Therefore, it is preferable to design a robust QSI signal which not only indicateswhether a group of UEs can be put to sleep quickly, but also helps in faster timingsynchronization. A UE decoding such a QSI signal would obtain both the paging andtiming information in parallel which could reduce the ON time and paging decodingcomplexity, thereby saving energy. In this section, we present the QSI signal designmechanism for MTC UEs with CE using dedicated resources in the PDSCH space.In particular, we propose to use ZC sequences to create QSI signals which possessgood auto-correlation and cross-correlation properties and thus enable robust signal45Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping. . . . .PSS and SSS QSI PDCCHSF 1 SF 2 SF 6 SF 7 SF 1 SF 2Figure 2.5: QSI transmission mechanism on PDSCH.detection. ZC sequences are Complex Amplitude Zero AutoCorrelation (CAZAC)sequences and cyclically shifted versions of these sequences are orthogonal to eachother [58]. Also, the cross-correlation of two ZC sequences of length N is limited by1√N. They are already being used in LTE/LTE-A for PSS and random access. Theproposed QSI ZC sequence is of the formQSIZC(p) = e(−jpipn(n+1)N ) (2.3)where, N = 131 is the length of the ZC sequence and p is the root of the ZCsequence chosen such that it is co-prime with N . We choose p ∈ [2 + 8 · (q − 1)] whereq = 1, 2, · · · 16. The QSI sequence occupies 131 subcarriers. This length is chosenbecause a legacy paging block would take at least 1 PRB pair and considering 2symbol-PDCCH, this would occupy 132 subcarriers [57,59]. One could always choosea longer length sequence to improve the performance since the cross-correlation peakis inversely proportional to the length of the sequence, but a longer sequence wouldrequire more resources.Our proposed QSI transmission mechanism in the PDSCH space uses 1 PRB pairin subframes 1, 2, 6 and 7 of a radio frame. These subframes are chosen to providetime diversity to the QSI signal and ensure that it is periodic. We use the centre1.4 MHz band and transmit the QSI ZC sequence on the top PRB pair of subframes1 and 7 and the bottom PRB pair of subframes 2 and 5, thereby introducing some46Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingfrequency diversity to the QSI signal. This also ensures that the QSI pattern has aperiodicity of 10 ms and the UE can determine the exact subframe number when itdetects the QSI. Figure 2.5 illustrates the proposed QSI signaling pattern.2.5 Energy Consumption and ComputationalComplexity AnalysisIn this section, we analyze the reduction in UE energy consumption and the numberof UE computations for our different QSI solutions. Prior to this work, the energyconsumption of the UE in DRX mode has been analyzed using a semi-Markov chainmodel [78, 88] or a queueing based model [80]. In our work, we use a simpler modelwhere the energy consumption is derived from the ON time of the UE similar to [89]and the computational complexity is determined using the number of FFT operationsperformed by the UE.2.5.1 Energy Consumption AnalysisLet tDRX, tSync and tPaging denote the total time of UE DRX cycle, time taken by theUE for synchronization and the UE ON time for paging decode respectively. Thetime spent by the UE in the “Light Sleep” state and the drift time of the UE clockare represented by tLS and tDrift respectively. PON, PLS and PDS denote the powerconsumed by the UE during the ON state, the “Light Sleep” state and the “DeepSleep” state respectively.47Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping2.5.1.1 Legacy UE in Idle Mode DRXFirstly, we consider the case of a legacy UE without CE at an operating SNR goodenough to decode PSS/SSS and PBCH in the first attempt. For PSS/SSS detection,we assume that the UE is ON for a duration of one subframe since the UE has tosearch for PSS/SSS within the buffered subframe, which gives tSync = 1 ms. When thepaging subframe is 0 or 5, the legacy UE does not wait for PO and tLS = 0. However,if the paging subframe is 4 or 9, the UE decodes the PSS/SSS on subframe 0 or onsubframe 5 and has to wait 3 more subframes for its PO. The UE goes into “LightSleep” for these subframes, which gives tLS = 3 ms. In the case of MTC for IoT, dueto the large number of UEs being paged, we can assume that the POs are equallylikely and compute the average light sleep time tavgLS =0+3+0+34= 1.25 ms. The totalON time of the legacy UE is given by tLegacyON = tDrift + tSync + tPaging. Therefore, theenergy consumed by the legacy UE can be calculated asELegacy = tLegacyON PON + tavgLS PLS + (tDRX − tLegacyON − tavgLS )PDS. (2.4)Secondly, we consider the case of a legacy UE with CE. In this case, the SNRis low and multiple repetitions are required for successful detection of PSS/SSS andpaging. Therefore, the total energy consumed by the legacy UE with CE can becalculated using Eq. (2.4) with tLegacyON = tDrift + tCESync + tCEPaging, where tCESync and tCEPagingdenote the ON time required for PSS/SSS detection and paging decode with CE,respectively. MTC UE without CE Adopting Our DRX with QSI mechanismHere, the QSI is transmitted on PBCH or on PSS/SSS. In both the cases, similar tothe legacy case, tSync = 1 ms, but the average light sleep time varies depending on48Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingwhere the QSI is transmitted and whether PBCH decoding is necessary. If the QSI istransmitted on PSS/SSS and PBCH decoding is not necessary, then the average lightsleep time is the same as that of the legacy UE, that is, tavgLS = 1.25 ms. However, ifthe QSI is transmitted on PBCH or if the UE has slept long enough so that PBCHdecoding is necessary, then the UE wakes up at subframe 0 regardless of the PO,detects PSS/SSS on subframe 0, decodes the PBCH, obtains the QSI and goes into“Light Sleep” until the PO subframe. Therefore, tLS can be 0 ms, 3 ms, 4 ms or8 ms and the average light sleep time tavgLS = 3.5 ms. Let p indicate the probability ofsuccessful QSI detection and q indicate the probability that the QSI conveys “sleep”.Then, the probability of successful QSI detection and UE going back to sleep is pqand the total ON time for the UE in this case will be tQSI = tDrift + tSync. Thecorresponding energy consumed by the UE isE1 = tQSIPON + (tDRX − tQSI)PDS. (2.5)If the QSI is detected successfully and the UE has to stay awake for paging or if theQSI signal is not detected, the UE resumes legacy mode of operation. This occurswith probability (1 − pq) and the energy consumed by the UE is ELegacy given by(2.4). Therefore, the total energy consumption of the UE without CE adopting QSIcan be calculated asEQSI = pqE1 + (1− pq)ELegacy. (2.6) MTC UE with CE Adopting Our DRX with QSI MechanismIn this case, the UE attempts to decode the QSI transmitted on PDSCH which givesthe UE both the timing information as well as sleeping indication. The QSI signaldetection, PSS/SSS detection and paging decode operations will take tQ, tCESync and49Chapter 2. Discontinuous Reception (DRX) with Quick SleepingtCEPaging amount of time respectively since the SNR is low and multiple repetitions arerequired for successful detection. The UE has to wake up at least tQ + tCESync beforethe PO since it has to first detect QSI and if it fails, the UE should fall back tolegacy operation, detect PSS/SSS followed by paging decode. The ON time for QSIdetection will be tCEQSI = tDrift + tQ. The energy consumed by the UE if it detects QSIsuccessfully and the QSI indicates “sleep” (which occurs with probability pq) is givenbyECE1 = tCEQSIPON + (tDRX − tCEQSI)PDS. (2.7)If the QSI is detected successfully and it indicates “stay-awake”, the UE goes intothe “Light Sleep” state until the PO subframe and then decodes the paging on thePO subframe. This occurs with probability p(1− q) and the energy consumed by theUE isECE2 = (tCEQSI + tCEPaging)PON + (tCESync + tavgLS )PLS + (tDRX− tCEQSI− tCEPaging− tCESync− tavgLS )PDS.(2.8)If the UE is unable to detect the QSI signal, then it resumes legacy operation andthis scenario occurs with probability (1− p). The energy consumed isECE3 = (tCEQSI+tCESync+tCEPaging)PON+tavgLS PLS+(tDRX−tCEQSI−tCESync−tCEPaging−tavgLS )PDS. (2.9)Therefore, the total energy consumption of the UE adopting QSI can be calculatedasECEQSI = pqECE1 + p(1− q)ECE2 + (1− p)ECE3 . (2.10)50Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping2.5.2 Computational Complexity AnalysisThe paging decoding requires PDCCH decoding as its first step regardless of whetherthe UE is being paged or not. PDCCH requires a full eNB bandwidth size FFT whichrequires O (N log2N) computations, N being the FFT size. Legacy UE in Idle Mode DRXFirst, we consider the case of a legacy UE without CE which takes one subframe forPSS/SSS detection to reacquire the timing and one subframe to decode the pagingblock. The PSS/SSS and the PBCH require a 128 point FFT [58]. Consideringthat for normal CP length, we have 14 symbols in a subframe, the number of FFToperations for synchronization is nSync = 14× 128 log2(128) = 12544. Assuming thatthe PDCCH occupies m symbols, the UE would require nPDCCH = m · O (N log2N)operations for PDCCH FFT. If the PDCCH indicates P-RNTI, the UE has to proceedto decode the PDSCH which will add to the number of FFT operations of the UE.For this analysis, we take into account only the synchronization and PDCCH FFToperations for the legacy UE since we model the scenario where a UE is very rarelypaged and the contribution of PDSCH FFT to the total number of FFT operations isnot significant. The total number of FFT operations for the legacy UE is computedasnLegacy = nSync + nPDCCH. (2.11)For the legacy UE with CE, the total number of FFT operations will benCELegacy = nSyncrSync + nPDCCHrPDCCH. (2.12)where rSync and rPDCCH is the number of PSS/SSS and PDCCH repetitions required51Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingfor successful decoding respectively. MTC UE without CE Adopting Our DRX with QSI MechanismHere, the QSI is sent on PBCH or on PSS/SSS and does not require additional FFTcompared to the legacy UE. The number of FFT operations for the UE withoutCE using QSI will be nSync with probability pq when QSI is detected and indicates“sleep”. Otherwise, the UE resumes legacy operation and number of FFT operationswill be equal to nLegacy. Therefore, the total number of FFT operations for the UEimplementing QSI on PBCH or QSI on PSS/SSS is given bynQSI = pqnSync + (1− pq)nLegacy. (2.13) MTC UE with CE Adopting Our DRX with QSI MechanismIn this case, the QSI signal is transmitted on PDSCH. The UE buffers the entire QSIsubframe and tries to detect the QSI signal. Since we transmit the QSI on 1 PRBpair in the PDSCH space, the minimum FFT size in LTE/LTE-A which is a 128 pointFFT, is sufficient to obtain the QSI signal. Thus, the number of FFT operations fordetecting the QSI signal can be calculated as nQ = 14×128 log2(128)×rQ = 12544rQ,where rQ is the number of repetitions required to decode the QSI signal successfully.If the QSI signal is detected successfully and if it indicates “sleep”, the number of FFToperations will be equal to nQ which occurs with a probability pq. If the QSI signal isdetected successfully and if it indicates “stay-awake”, it will be nQ + nPDCCHrPDCCHwhich happens with probability p(1− q). If the QSI is not detected, then the numberof FFT operations will be nQ+nCELegacy. Therefore, the total number of FFT operations52Chapter 2. Discontinuous Reception (DRX) with Quick Sleepingfor UE with CE adopting QSI isnCEQSI = pqnQ + p(1− q)(nQ + nPDCCHrPDCCH) + (1− p)(nQ + nCELegacy). (2.14)2.6 Simulation Results and AnalysisIn this section, we highlight the benefits of the proposed QSI solutions for LTE/LTE-A for IoT through simulation results and detailed analysis of the UE energy efficiencyand computational complexity when our QSI mechanisms are used.2.6.1 Simulation ResultsFirst, the methods discussed in Section 2.3.1 were implemented and simulations wererun for an Additive White Gaussian Noise (AWGN) channel. The AWGN channel waschosen since it is sufficient to determine performance differences for QSI and PBCHfor the different proposed methods. The key simulation parameters are summarizedin Table 2.2.Figure 2.6 shows the PBCH Block Error Rate (BLER) and also indicates theexpected BLER curves when QSI is introduced. The SNR denotes the SNR on eachPBCH symbol. The PBCH symbol also contains the cell specific reference signalsand the QSI signal. The expected BLER curves are obtained with the assumptionthat QSI does not interfere with the PBCH signal. and the loss in PBCH detectionperformance is (1− P )SNR as discussed in Section 2.3.1.It can be observed that the QSI using orthogonal sequences shows 0.22 dB degra-dation in PBCH BLER performance when compared to the legacy UE PBCH BLERperformance which matches the analytical results. However, there is a small differencebetween the expected BLER and actual BLER when the QSI signal is not orthogonal53Chapter 2. Discontinuous Reception (DRX) with Quick SleepingTable 2.2: Simulation parameterseNB Parameters ValueAntenna Configuration 2Tx × 1RxNumber of downlink RBs 6PHICH duration NormalPHICH Group Multiplier 16System Frame Number randi([0 255])Cell ID 0No. PDCCH symbols, m 2UE Parameters ValueAntenna Configuration 1Tx × 1RxVCO accuracy 10ppmDRX cycle length, [1.28 s, 2.56 s,tDRX 10.24 s, 1 min, 10 min]Drift time, VCO accuracy ×tDrift tDRXON time for paging decode, 1 ms (normal coverage)tPaging 10 ms (CE mode)ON state power, PON 500 mW“Light Sleep” power, PLS 250 mW“Deep Sleep” power, PDS 0.0185 mWProbability of successful 0.9QSI detection, pProbability that QSI 0.9indicates “sleep”, qto the PBCH signal, i.e., for the methods using spreading and repetition or repeti-tion plus FEC. The non-orthogonality of the QSI signal leads to interference withthe PBCH signal, which in turn leads to degradation in PBCH BLER performance.Figure 2.7 indicates the QSI BLER performance. As expected, QSI using orthog-onal codes gives the best performance. It would be preferred if the MTC UE canafford to have significant amount of memory to store the different orthogonal QSIsequences. From an implementation perspective, the MTC UEs have a high speed,high cost working memory space and a low speed, low cost permanent memory space.The UEs can store the orthogonal QSI sequences for all the Cell IDs in the perma-54Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping−6 −5 −4 −3 −2 −1 010−210−1SNR in dBPBCH Block Error RatePBCH Block Error Rate v/s SNR with P = 0.05  QSI with Orthogonal Codes (For all MIBs per Cell Id)QSI with Repetition + exp(−j*2*pi*n/SPF) spreadingQSI with Repetition + (8,4) Hamming code + exp(−j*2*pi*n/SPF) spreadingReference BLER without QSI (Legacy)Expected BLER for P = 0.05Figure 2.6: PBCH BLER performance for AWGN channel.nent memory and can have only the sequences corresponding to the detected Cell IDin the working memory so that the low cost feature of the MTC UE is not largelycompromised.It can be seen that QSI methods using repetition coding only and repetitioncoding with FEC also demonstrate good QSI BLER performance. These methodsare straight-forward to implement and unlike the QSI mechanism using orthogonalsequences, these methods do not require extra memory. However, there is about0.9 dB degradation in PBCH BLER performance at 1% BLER when compared tothe legacy UE PBCH BLER performance considering that the total power availablefor PBCH is unaltered. The degradation will not occur if the eNB can afford extrapower for QSI transmission.Next, the different QSI solutions were simulated using the LTE system toolboxon MATLAB for the Extended Pedestrian A (EPA) channel model with a DopplerSpread of 1 Hz and the number of QSI bits, M = 4. This channel model was selected55Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping−6 −5 −4 −3 −2 −1 010−510−410−310−210−1100SNR in dBQSI Block Error RateQSI Block Error Rate v/s SNR with P = 0.05  QSI with Orthogonal Codes (For all MIBs per Cell Id)QSI with Repetition  + exp(−j*2*pi*n/SPF) spreadingQSI with Repetition + (8,4) Hamming code + exp(−j*2*pi*n/SPF) spreadingFigure 2.7: QSI BLER performance for AWGN channel.since it is recommended by the 3GPP for testing low UE mobility [106] and is suitablefor the IoT scenarios considered in this work such as pet tracking and weather sensingin the case of normal coverage or patient monitoring in case of enhanced coverage.Other simulation parameters are the same as in Table 2.2.The QSI solutions for MTC UEs without CE discussed in Section. 2.3 were ana-lyzed using the BLER performance considering the minimum SNR required for 10%BLER (S10%) as the performance indicator. For MTC UEs with CE, we consid-ered 15 dB CE and obtained the operating SNR for the synchronization channel as−14.2 dB [43]. We evaluated the performance based on the accumulation time re-quired for 90% detection (tAcc) of the desired signal (PSS/SSS for the legacy UE andthe QSI signal for UE with QSI) for false alarm (FA) rates of 10% and 1%. Table 2.3summarizes the performance results for the different QSI solutions.The results for QSI on PBCH and QSI on PSS/SSS are shown in Figure 2.8aand Figure 2.8b, respectively. For QSI on PBCH, method 1 corresponds to spreading56Chapter 2. Discontinuous Reception (DRX) with Quick SleepingSNR in dB-10 -8 -6 -4 -2 0 2 4 6BLER10 -310 -210 -110 0QSI on PBCH BLER with P = 0.05Legacy PBCH (No QSI)Expected BLER for PBCH + QSIPBCH with QSI Method 1QSI Method 1PBCH with QSI Method 2QSI Method 2(a) QSI on PBCHSNR in dB-10 -8 -6 -4 -2 0 2 4 6BLER10 -310 -210 -110 0QSI on PSS/SSS BLERPSS/SSS Legacy (no QSI)PSS/SSS with QSIExpected BLER for PSS/SSS + QSIQSI Method 1QSI Method 2QSI Method 3QSI Method 4(b) QSI on PSS/SSSFigure 2.8: Performance results for QSI without CE using the EPA channel model.57Chapter 2. Discontinuous Reception (DRX) with Quick SleepingTable 2.3: Performance summary using S10% (without CE) and tAcc (with CE)PBCH QSI on PSS/SSS QSI on Legacy QSI onS10% PBCH S10% PSS/SSS PSS/SSS PDSCH(No CE) (No CE) (No CE) S10% tAcc tAcc(No CE) (CE) (CE)Legacy: 0.35 dB Met 1: Legacy: Met 1: 4.79 dB FA 10%: FA 10%:With QSI: 2.43 dB -4.41 dB Met 2: 2.37 dB 90 ms 20 msMet 1: 0.53 dB Met 2: With QSI: Met 3: 3.62 dB FA 1%: FA 1%:Met 2: 0.55 dB 1.33 dB -3.62 dB Met 4: 1.85 dB 200 ms 50 msQSI using repetition only (see Section and method 2 corresponds to spreadingQSI using repetition and FEC (see Section From Table 2.3, we note thatthe two QSI on PBCH solutions show 0.18 dB and 0.2 dB degradation in PBCHdetection performance respectively for 10% BLER, which is close to the expected losscalculated as 10 log10(1−P ) = 0.22 dB with P = 0.05 in Section 2.3. For the AWGNchannel results, the QSI signal using our proposed methods acted as interferenceto the PBCH signal and degraded the PBCH detection performance by more than0.5 dB. The same is true in the current simulation too, but the EPA channel is nota static channel like AWGN and the degradation in PBCH detection performance isnot as pronounced as it was for AWGN. For QSI on PSS/SSS, the expected loss inPSS/SSS detection performance when QSI is transmitted on the unused subcarriers is0.54 dB as computed in Section 2.3. From Table 2.3, we observe that the loss obtainedfrom our simulation is 0.79 dB for 10% BLER, which is close to our expected results.We observe that for QSI on PBCH using spreading, repetition and (8,4) extendedHamming code (method 2) gives the best performance and for QSI on PSS/SSS, themethod of transmitting QSI using the unused carriers of both PSS and SSS along with(8,4) extended Hamming code (method 4) gives the best performance. Comparingthe S10% for these two methods, we can see that QSI on PBCH (method 2) is 0.52 dBbetter than QSI on PSS/SSS (method 4) owing to the SNR gain due to spreading.58Chapter 2. Discontinuous Reception (DRX) with Quick SleepingAccumulation Time (ms)50 100 150 200 250 300 350 400 450 500Probability of detection0. and QSI performance at SNR = -14.2dBPSS/SSS Detection with 10% False AlarmQSI Detection with 10% False AlarmPSS/SSS Detection with 1% False AlarmQSI Detection with 1% False AlarmFigure 2.9: Re-Sync and QSI on PDSCH for CE using the EPA channel model.In the case of MTC UEs with CE, we determined the number of PSS/SSS repeti-tions required to reacquire the symbol timing when the UE wakes up for the legacyUE. Since we look at the case where the UE had already obtained the cell identi-fier before it went to sleep, the PSS/SSS is known to the UE. Therefore, the UEneed not hypothesize over all combinations of PSS/SSS and has to re-acquire thesame PSS/SSS combination which it detected before. Since the SSS and PSS are onconsecutive symbols, we considered them as one long sequence and used differentialauto-correlation to detect this long sequence and obtain the symbol boundary. A suc-cessful detection is registered when we detect the correct symbol boundary, otherwiseit is regarded as a false alarm. Similarly, we obtained the QSI detection performancewith the QSI transmitted on PDSCH. Figure 2.9 indicates the re-sync performanceof the legacy UE and the QSI detection performance for MTC UE with CE. Usingthe 15dB CE results summarized in Table 2.3, we observe that for the legacy UErequires tAcc = 200 ms (40 PSS/SSS repetitions), while the UE adopting our QSI onPDSCH mechanism requires tAcc = 50 ms (20 QSI repetitions) at a false alarm rate59Chapter 2. Discontinuous Reception (DRX) with Quick SleepingTable 2.4: Reduction in energy consumption for UE with QSIDRX QSI on QSI on QSI onlength PBCH PSS/SSS PDSCH(No CE) (No CE) (15 dB CE)1.28 s 45.67% 51.74% 63.88%2.56 s 44.67% 50.61% 63.86%10.24 s 39.48% 44.73% 63.75%1 min 22.52% 25.52% 63.05%10 min 7.36% 7.36% 56.49%of 1%. Therefore, the re-acquisition time of the UE is reduced by a factor of 4 atthis false alarm rate when our QSI mechanism is adopted, which improves the energyefficiency.2.6.2 Energy Efficiency ResultsThe energy efficiency is calculated as the ratio of the energy consumed by the UEusing our QSI solution to the energy consumed by the legacy UE. The energy con-sumed by the legacy UE was computed using (2.4) and the energy consumed bythe UE using QSI was computed using (2.6) or (2.10) depending on whether it is innormal coverage mode or coverage enhancement mode with the parameters listed inTable 2.2. For QSI on PDSCH, we considered 15 dB CE and used tQ = 50 ms andtCESync = 200 ms corresponding to tAcc at a false alarm rate of 1%.Table 2.4 summarizes the reduction in energy consumption obtained by using QSIfor different DRX cycle lengths. With the VCO drift of 10 ppm, the UE would requireto decode the PBCH if it sleeps more than 8.2 minutes. When PBCH decoding isnot required, QSI on PSS/SSS is more energy efficient than QSI on PBCH for MTCUEs without CE. If PBCH decoding is required, the energy efficiency obtained byboth the QSI mechanisms is equivalent. For MTC UEs with CE, QSI on PDSCHdemonstrates considerable improvement in energy efficiency since the ON time of the60Chapter 2. Discontinuous Reception (DRX) with Quick SleepingUE is reduced significantly. The energy efficiency is obtained by the reduction in theON time of the UE which depends on the SNR and not on the length of the DRXcycle. Therefore, at a given SNR, the ratio of ON time to the DRX cycle lengthdecreases with increasing DRX cycle length which leads to a decrease in the energyefficiency.2.6.3 Computational Complexity ResultsThe computational reduction is calculated as the ratio of the number of FFT com-putations consumed by the UE using our QSI solutions to that of the legacy UE.We obtained the number of FFT computations for the legacy UE using Eq. (2.11)for normal mode and Eq. (2.12) for coverage enhancement mode. For the UE us-ing our QSI solutions, we obtained the numbers from Eq. (2.13) and Eq. (2.14) fornormal mode and CE mode respectively. The number of repetitions, rSync, rQ andrPDCCH, were chosen corresponding to tSync, tQSI and tPaging respectively. The com-putational reduction obtained for different QSI solutions is summarized in Table 2.5.The PDCCH FFT size is directly proportional to the eNB bandwidth while QSI FFTsize is always 128 point. Therefore, the computational efficiency increases for highereNB bandwidths. This is true even for the CE case. Additionally, since the QSIdetection requires a lesser number of repetitions than legacy PSS/SSS detection, thecomputational efficiency obtained for the CE case is higher than that for the non-CEcase.It should be noted that along with the full scale FFT, PDCCH message blockdecoding also requires Viterbi decoding and hypothesizing over 44 different possiblelocations, which is computationally intensive, but the QSI either uses correlation with2M different sequences or uses simple despreading and repetition combining (plus ML61Chapter 2. Discontinuous Reception (DRX) with Quick SleepingTable 2.5: FFT computation reduction for UE with QSIeNB FFT QSI on PBCH QSI onBandwidth Size or PSS/SSS PDSCH(No CE) (No CE) (15 dB CE)1.4 MHz 128 10.69% 65.15%5 MHz 512 36.21% 65.73%10 MHz 1024 53.02% 66.55%20 MHz 2048 66.88% 68.12%decoding for 2M sequences in case of FEC), which are simpler methods compared toPDCCH blind decoding. This leads to further improvement in computational savings.2.7 ConclusionIn this chapter, we considered the problem of improving the current DRX and pagingmechanism for energy efficient IoT using LTE/LTE-A. We proposed a modified DRXmechanism incorporating quick sleeping as a novel, simple and efficient solution forthis problem. For the MTC UEs in normal coverage, we proposed QSI on PBCHand QSI on PSS/SSS mechanisms which do not require additional resources. ForUEs requiring extended coverage, we introduced the QSI mechanism using dedicatedresources on PDSCH. The different QSI solutions were simulated on the EPA-1 Hzchannel model to address the case of low-mobility. We also determined the reductionin energy consumption and computational complexity for the MTC UEs using ourQSI mechanisms when compared to the legacy UEs. For MTC UEs without CE usingour QSI solutions, we showed 45% and 66% reduction in energy consumption andcomputational complexity respectively. For MTC UEs with 15 dB CE using QSI onPDSCH, we demonstrated that we could obtain 63% reduction in energy consumptionand 68% reduction in computational complexity.Our QSI solutions are in line with the standardization activities for MTC in62Chapter 2. Discontinuous Reception (DRX) with Quick Sleeping3GPP LTE/LTE-A to facilitate IoT and have minimal influence on the legacy UEs.The specific QSI solutions provided in this chapter are suitable for CAT-M1, CAT-0and above UE categories, since they operate on a bandwidth of at least 1.4 MHz.However, the main idea behind the QSI, i.e., simplifying the paging decode processwhen there is no valid page and reducing the resynchronization time of MTC UEs, isapplicable to all UE categories and would essentially aid the design of a more efficientpaging and power saving mechanisms in the downlink for the forthcoming 5G NRstandard.63Chapter 3Enhanced PrimarySynchronization Signal (ePSS)3.1 IntroductionIn a typical wireless communication system, it is important for the mobile device, orthe so called UE to maintain accurate symbol timing synchronization with the basestation in order to decode the downlink data. In the previous chapter, the timingresolution considered was at the symbol level, i.e., the timing was designated to becorrect if the MTC UE supporting the IoT finds the correct symbol number. In thischapter, we examine the sensitivity of the paging decode operation to the timingoffset and show that any deviation in timing beyond the CP length considerablydegrades the decoding process.Conventionally, in OFDM based systems, the UEs reacquire the timing usingCP autocorrelation [63,64] and/or by detecting the synchronization reference signals[107–110] which are transmitted periodically by the base station. In this chapter,we show that the legacy methods for resynchronization using CP autocorrelationand reference signal detection are not effective for the Low-Cost, Low-Complexity,Low-Coverage (LC) MTC devices due to the increased ON time of the UE.For faster timing reacquisition during the DRX wake-up interval, we introduceour novel enhanced Primary Synchronization Signal (ePSS) as the resynchronization64Chapter 3. Enhanced Primary Synchronization Signal (ePSS)signal for the LC devices and demonstrate the reduction in energy consumption whenthe ePSS is used. Further, we also adopt DRX with QSI mechanism from Chapter 2and illustrate that using the ePSS as QSI combines the advantages of faster timingreacquisition and quicker transition to sleep mode when there is no page, begettingfurther improvement in the energy efficiency of the LC devices.As before, we ensure that our solutions are designed such that the changes requiredto the current LTE/LTE-A standards are kept minimal and the resource allocationrespects the procedure followed in the current standards.The rest of the chapter is outlined as follows. In Section 3.2, we demonstrate theimportance of timing reacquisition for the LC devices and explore the conventionalmechanisms for timing reacquisition. In Section 3.3, we describe the design and re-source allocation aspects of our proposed ePSS mechanism and discuss how the ePSScan also be used as QSI to reduce the ON time of the MTC UE. In Section 3.4, wecompare the detection performance of our ePSS signal and the legacy synchroniza-tion signals (PSS/SSS) and illustrate the energy efficiency obtained when our ePSSis adopted by the LC devices in DRX. The conclusions are drawn in Section UE Timing Accuracy and Legacy TimingAcquisition AlgorithmsAs discussed in Section 1.2.1, the paging message consists of the P-RNTI on thePDCCH, followed by the paging data on the PDSCH. The performance of the PDCCHand the PDSCH has been analyzed in many prior works. For example, in [111],the PDCCH performance is analyzed for M2M traffic and the scenario of excessload on the PDCCH due to large number of MTC devices is discussed. In [112]65Chapter 3. Enhanced Primary Synchronization Signal (ePSS)and [113], the PDCCH BLER performance is analyzed for various channel modelsand the PDSCH BLER performance is discussed in [114]. These works assume perfecttiming synchronization. However, the LC devices waking up to decode the pagingmessage will have a clock drift depending on the quality of the oscillator used for theUE’s clock.If the UE uses a high quality oscillator, the timing drift will be small. Thismeans that the UE can sleep for a longer duration without losing the timing syn-chronization. Typically, mobile devices like smartphones using a high quality VoltageControlled Temperature Compensated Crystal Oscillator (VCTCXO) have an accu-racy of ±1 ppm [115]. But, the low-complexity MTC UEs cannot incorporate ahigh quality oscillator for its clock since it increases the cost of the device. Hence,most of them use a VCO for their clock, which has an accuracy of ±10 ppm [116].As an example, assuming that the symbol time is 72 µs and the tolerable timingdrift is 5%, i.e. 3.6 µs, the MTC UE with a 10 ppm accurate clock can sleep up to3.6µs10.10−6 = 0.36 s= 360 ms, while the device with a 1 ppm accurate clock can sleep upto 3.6µs1.10−6 = 3.6 s. However, the sleep time supported by the network can be longer.For example, in LTE/LTE-A, the sleep time supported by the network has beenrecently extended to 2621.44 s (43.69 minutes) [56]. Therefore, the assumption ofperfect timing synchronization is not always true for both types of devices and tim-ing reacquisition becomes important for successful communication between the UEand the eNB.In this section, we first demonstrate the sensitivity of the PDCCH/PDSCH pagingdecode operation to the UE timing offset. Then, we explore the timing estimation andsynchronization algorithms used widely in OFDM based systems. We examine theperformance of two algorithms - a) CP autocorrelation [63,64] and b) Synchronization66Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Table 3.1: Simulation settings for Figure 3.1.Parameter SettingDownlink Bandwidth 1.4 MHzSampling Rate 1.92 MHzAntenna Configuration 2× 1Channel Model EPAMax. Doppler Shift 1 HzMIMO Correlation LowType = Format 2PDCCH Settings RNTI used = P-RNTIDCI Format = 1AMCS = 0PDSCH Settings TBS = 16 bitsNumber of PRBs = 1signal detection [107–110] and determine their suitability for low-complexity MTCUEs in CE mode. The LTE toolbox in MATLAB was used for the simulations. TheEPA 1 Hz channel was chosen since it is the recommended channel model to studythe performance of MTC in LTE/LTE-A [43,106] because it models the low mobilitymulti-path scenario quite well. Especially for the MTC UEs in low coverage, the3GPP envisions the scenario where a significant portion of these devices are locatedin basements of buildings or underground parking lots and are mostly stationary,which is well characterized by the EPA 1 Hz channel model [43, 106]. The CFOvalue for timing reacquisition was chosen to be 1 kHz [43]. But for the PDCCHand the PDSCH decoding, the CFO was set to 100 Hz, since the initial CFO will beestimated and compensated after timing reacquisition and one has to account only forthe residual CFO [43]. The other important simulation parameters are summarizedin Table 3.1.67Chapter 3. Enhanced Primary Synchronization Signal (ePSS)SNR in dB-15 -10 -5 0 5 10BLER10 -310 -210 -110 0TO = 0TO = 7 samples = 5.11%TO = 10 samples = 7.3%(a) PDCCH BLER performanceSNR in dB-15 -10 -5 0 5 10BLER10 -310 -210 -110 0TO = 0TO = 7 samples = 5.11%TO = 10 samples = 7.3%(b) PDSCH BLER performanceFigure 3.1: Illustration of the sensitivity of the PDCCH and the PDSCH decodingto timing offset (TO).68Chapter 3. Enhanced Primary Synchronization Signal (ePSS)3.2.1 Importance of UE Timing AccuracyThe first step of the paging decode process is to check for the P-RNTI by decoding thePDCCH. If the P-RNTI is found, the UE decodes the PDSCH to check for the pagingdata. Otherwise, it goes back to sleep. It is important to analyze the sensitivity of thePDCCH and the PDSCH performance with respect to UE timing accuracy. Similarto the previous works in [113, 114], we use the BLER as the metric for analyzingthe performance the PDCCH containing the P-RNTI and a PDSCH block containingpaging data.Figure 3.1a and Figure 3.1b demonstrate the BLER performance of the PDCCHand the PDSCH respectively. It is evident that both the PDCCH and the PDSCHare highly sensitive to timing offset. We consider the 10% BLER point to analyze theperformance as suggested in [43] for LTE/LTE-A MTC. At 10% BLER, the PDCCHperformance is degraded by about 8 dB for as little as 7.5% error in timing estimationand the corresponding degradation in the PDSCH performance is 12 dB. However,when the timing estimate is within 5% of the actual value, the degradation is lessthan 3 dB in both the cases. Therefore, the accuracy of UE timing plays a vitalrole in the success of the paging decode mechanism in LTE/LTE-A. Additionally,for the LC devices, the timing reacquisition algorithms would be required to provideaccurate results at very low SNR. Next, we discuss the performance of legacy timingestimation algorithms for the LC devices operating at low SNRs.3.2.2 Timing Reacquisition Using CP AutocorrelationIn this method, the symbol start position is obtained by finding the location of thepeak of the CP autocorrelation [58, 63, 64]. For the illustration of the performanceof CP autocorrelation, we accumulate the correlation results only on the symbols69Chapter 3. Enhanced Primary Synchronization Signal (ePSS)SNR in dB-20 -15 -10 -5 0 5Probability of Error in Timing Detection10 -310 -210 -110 010ms100ms400ms600ms800msFigure 3.2: CP autocorrelation performance.that are used to transmit the pilot signal. This is because the pilot signal is alwaystransmitted on specific symbols regardless of whether the eNB has data or controlinformation to send. However, on the other symbols, the subcarriers contain a validsignal only when they are used for control or data allocation. The pilot signal istransmitted on the first and fifth symbol of each slot (see Figure 1.2a) [58,61].Figure 3.2 depicts the performance of CP autocorrelation for different accumula-tion times. For the LC devices, the operating SNR for the synchronization channelat 10% BLER is -14.2 dB [43] and the error tolerance level of the estimated timingoffset is set to ±5% based on the results obtained in Section 3.2.1. From Figure 3.2,we observe that about 600 subframes (corresponding to 600 ms) of accumulation timeis required for 10% BLER at -14.2 dB which is large. This is because there are only9 CP samples available for correlation at a sampling rate of 1.92 MHz (the samplingrate supported by the low-complexity MTC UEs [43]). This means that the MTCUE has to be on for 600 ms, which will lead to increased energy consumption. There-fore, we conclude that the classical method of CP autocorrelation is not an efficient70Chapter 3. Enhanced Primary Synchronization Signal (ePSS)SNR in dB-15 -10 -5 0 5Probability of Error in Timing Detection10 -210 -110 010ms50ms100ms200ms300ms400ms500msFigure 3.3: PSS/SSS detection performance.mechanism for timing reacquisition for low-complexity MTC UEs in CE mode.3.2.3 Timing Reacquisition Using Synchronization SignalDetectionFor the synchronization signal detection, we consider the PSS and the SSS as one longsequence, since they are on consecutive symbols. For successful timing acquisition,we have to find the position of this sequence in a 10 ms radio frame within a tolerancelevel (set to ±5% in our implementation similar to Section 3.2.2). We use differentialautocorrelation in frequency domain to detect the synchronization signal [107,108].Figure 3.3 gives the performance results for timing reacquisition using the legacysynchronization signal detection. Considering 10% BLER at the coverage enhance-ment SNR of -14.2 dB [43], we require around 400 ms, i.e., 80 repetitions of the legacy71Chapter 3. Enhanced Primary Synchronization Signal (ePSS)synchronization signal to detect the correct timing offset. This is again significant,resulting in increased energy consumption of the UE. Therefore, we require a fasterresynchronization mechanism to improve the energy efficiency of the LC devices.3.3 Technicalities of the ePSSIn this section, we introduce our new, simple resynchronization signal specially cater-ing the needs of the LC devices. Since this signal is designed similar to the PSS, itwould be apt to call it the enhanced PSS (ePSS). We discuss the design and resourceallocation aspects of our ePSS mechanism.3.3.1 ePSS DesignThe ePSS signal should be designed such that the MTC UE can detect it with con-siderable accuracy at very low SNRs (around -14 dB). Since the ePSS is used by LCdevices, which have limited processing capabilities, it should be designed to providerobust detection with minimal complexity. This requires the ePSS signal to possessgood autocorrelation and crosscorrelation properties. Such properties are demon-strated by the ZC sequences that are extensively used in the LTE/LTE-A standards,for example, in the case of the PSS in the downlink and the Physical Random AccessChannel (PRACH) in the uplink. The ZC sequences are such that their cyclicallyshifted versions are orthogonal to each other and the crosscorrelation of two N lengthZC sequences is limited by 1√N[58]. These properties make them the perfect candi-dates for the ePSS signal design.Also, the location of the ePSS in the LTE/LTE-A radio frame should be suchthat the UE can unambiguously determine the subframe number upon successfuldetection, which necessitates the ePSS to occupy exclusive and invariable resources72Chapter 3. Enhanced Primary Synchronization Signal (ePSS)SF 1SF 0Slot 00.5msSF 2 SF 3ePSS1ePSS1ePSS1ePSS1ePSS1ePSS10.5msSlot 1 c  control symbols d  data symbolsePSS2ePSS2ePSS2ePSS2ePSS2ePSS2For normal CP, c = 2, d = 12For extended CP, c = 1, d = 11(a) ePSS subframe structure12 REs PSS Root r0PSS Root r1PSS Root r1PSS Root r2ePSS 131-length ZC Root r0ePSS 131-length ZC Root r1ePSS1 ePSS2ePSS1 ePSS2ePSS Using PSS ZCePSS Using Longer ZCd = 12 symbolsd = 12 symbols12 REs(b) Normal CP - Using only data symbolsPSS Root r0PSS Root r1 ePSS 131-length ZC Root r0ePSS1 ePSS2ePSS1 ePSS2ePSS Using PSS ZCePSS Using Longer ZCPSS Root r1PSS Root r2 ePSS 131-length ZC Root r1c + d = 12 symbols12 REs12 REsc + d = 12 symbols(c) Extended CP - Using both control and datasymbolsPSS Root r0(1:60)PSS Root r1(1:60) ePSS 119-length ZC Root r0ePSS1 ePSS2ePSS1 ePSS2ePSS Using PSS ZCePSS Using Longer ZCPSS Root r1(1:60)PSS Root r2(1:60) ePSS 119-length ZC Root r1d = 11 symbols12 REsd = 11 symbols12 REs(d) Extended CP - Using only data symbolsFigure 3.4: Illustration of ePSS subframe structure for normal and extended time and frequency. In order to obey the mandate of resource allocation dictated bythe current LTE/LTE-A framework, such dedicated resources can be accommodatedonly in the PDSCH space. The number of symbols available for the PDSCH is decidedby the eNB.73Chapter 3. Enhanced Primary Synchronization Signal (ePSS)We use the PDSCH on subframes 1 and 2 for transmitting the ePSS, so that itslocation is fixed in time as shown in Figure 3.4a. In LTE/LTE-A with normal CP,a subframe consists of 14 symbols. A Resource Element (RE) spans 1 subcarrier ×1 symbol and a PRB consists of 12 REs × 7 symbols = 84 REs. The minimum unitof allocation spans 12 REs × 1 subframe, which corresponds to a pair of PRBs =168 REs. A commonly used configuration for LTE/LTE-A MTC using normal CP isa subframe consisting of a 12 symbol PDSCH preceded by a 2 symbol PDCCH [43].In this scenario, there are 168 - 2 × 12 = 144 REs per PRB pair available for thePDSCH. Also, some REs in the PDSCH are reserved for pilot signals, which is 12REs per PRB pair in this case. This gives us 132 REs per PRB pair for the PDSCH.In the extended CP case, a subframe consists of 12 symbols and the minimumunit of allocation spans 12 REs × 1 subframe = 12 × 12 = 144 REs. As in the case ofnormal CP, 12 REs per PRB pair are required for pilot signal transmission. Hence,the total number of REs available per PRB pair is 144-12 = 132 REs. Typically, onesymbol is used for the PDCCH [43]. Therefore, we have 132 - 12 = 120 REs availablefor the PDSCH for the extended CP case.In the following, we present two methods to design the ePSS - a) Using multiplePSS and b) Using longer ZC sequences. The ePSS detection uses differential autocor-relation similar to the legacy synchronization signal detection mechanism discussedin Section ePSS Construction Using Multiple PSSIn this method, the ePSS is formed by a burst of PSS copies occupying the REs inthe PDSCH space. The advantage of re-using the existing PSS sequences is that theyare readily available at the eNB and no additional processing/memory is requiredto generate/store a new sequence. An example for such an ePSS for normal CP74Chapter 3. Enhanced Primary Synchronization Signal (ePSS)is shown in Figure 3.4b. An ePSS PRB consists of the concatenations of two PSSsequences of different roots. Unlike the regular PSS where the ZC sequence is spreadacross frequency, the PSS ZC sequence within our ePSS is spread across time (seeFigure 3.4b). This ensures that the legacy UEs do not falsely detect this signal asthe PSS. The PSS signal is a 63-length ZC sequence [58] and two such sequences willoccupy 126 REs. The unused REs are set to zero. The low-complexity MTC UEscan operate on a maximum downlink bandwidth of 1.4 MHz which corresponds to 6PRBs. Therefore, we transmit 6 copies of the ePSS in a subframe which is equivalentto transmitting 12 PSS copies.In the ePSS design, we cannot have the same signal on both subframe 1 andsubframe 2, since the ePSS will be used for timing resynchronization. If these twosubframes have the same signal, the UE would not be able to uniquely determine thesubframe number. Therefore, we have to judiciously re-use the PSS sequences so thatthe two subframes are different, but can be detected together to uniquely determinethe subframe number. The LTE/LTE-A standards use 3 roots for the PSS sequence- r ∈ [r0 = 25, r1 = 29, r2 = 34]. Each ePSS sequence consists of 2 of the 3 possiblePSS roots. This gives us 6 possible ways to choose the roots for subframe 1. Theyare (r0, r1), (r0, r2), (r1, r0), (r1, r2), (r2, r0) and (r2, r1).Let us say that subframe 1 has the sequence (r0, r1) and call r0 the top root andr1 the bottom root. Since the ePSS is used for timing reacquisition, if both thesubframes contain the same top and bottom roots, the UE using the ePSS to detectthe timing will not know whether it detected the root on subframe 1 or on subframe 2.Therefore, we should have different sequences on subframe 1 and subframe 2. Hence,subfame 2 can only have (r1, r0), (r1, r2) or (r2, r0) as the sequence. Similarly, if westart with a different sequence for subframe 1, we have 3 possibilities for subframe75Chapter 3. Enhanced Primary Synchronization Signal (ePSS)2. Thus, in total we have 6 × 3 = 18 sequences to construct the ePSS using thismethod.Moreover, there are only 3 legacy PSS sequences and one sequence is used persector. Therefore, the frequency reuse for the legacy scheme is 3, which may resultin the detection of the neighbour PSS. However, for ePSS constructed using thismethod, we have 18 possible sequences (greater than 7). Therefore, the ePSS canbe used with a frequency re-use factor of 7, which reduces the probability of theneighbour ePSS detection.For the extended CP case, the ePSS transmission uses one of the following schemes- a) using both the control and data symbols (as shown in Figure 3.4c) and b) usingonly the data symbols (as shown in Figure 3.4d). In the first scheme, there are 132REs available per PRB pair (since the control symbols are also used) and the ePSSsequences are the same as that for normal CP. However in the second scheme, thereare 120 REs available per PRB pair and the ePSS sequences are slightly shortenedso that they can be accommodated within the available space with negligible loss inperformance. ePSS Construction Using Different ZC SequencesIn this method, we use ZC sequences for the ePSS which are different from thoseused for the PSS. The ZC sequence is of the form ZC(r) = e(−jpirn(n+1)N ), where n =0, 1, · · ·N − 1, r is the root of the ZC sequence and N is the length. The root rand the length N are co-prime. As discussed earlier, for the PSS, N = 63 andr ∈ [25, 29, 34]. There are 33 more roots that are co-prime to 63 which can be usedto construct different ZC sequences. The ePSS can be then constructed as describedin Section second solution for using different ZC sequences consists of using a longer76Chapter 3. Enhanced Primary Synchronization Signal (ePSS)length sequence for the ePSS. For example, in our scenario of using a PDSCH with132 REs per PRB, we can use a ZC sequence of length 131 and set the single unusedRE to zero. Also, using a longer length sequence provides a larger set of sequencessince the number of roots co-prime with the length increases. Moreover, the cross-correlation between two ZC sequences is proportional to 1√Nand a longer lengthsequence should improve the performance. Figure 3.4b also depicts the constructionof the ePSS using longer length ZC sequences for the normal CP case. Figure 3.4cand Figure 3.4d demonstrate the two ePSS transmission schemes for the extendedCP case, where the former scheme uses the same 131-length ZC sequences as that ofthe normal CP and the latter uses a 119-length shortened ZC sequences.Similar to Section, the signals on subframe 1 and subframe 2 should bedifferent so that the MTC UE can clearly identify the subframe number on resyn-chronization. With a 131 length sequence, we have 130 possible roots and 130 × 129= 16,770 sequences. This gives us a large set of sequences to choose for the ePSSsignal design. Again, such an ePSS can also be used with a frequency re-use factorof 7 due to the availability of multiple sequences, which reduces the probability ofdetecting a neighbour ePSS.3.3.2 ePSS AllocationThe ePSS will be used for resynchronization of the MTC UE with the same cell.Hence, the location(s) of the ePSS in the LTE frame structure should be known bythe UE beforehand. Therefore, the ePSS cannot be scheduled using the PDCCH orthe enhanced PDCCH (ePDCCH) [117]. We propose the following allocation schemes.• ePSS on centre band: This is a fixed allocation scheme where the REs usedfor the ePSS always correspond to the centre band occupying 1.4 MHz. The77Chapter 3. Enhanced Primary Synchronization Signal (ePSS)location of the ePSS in time is already fixed, i.e., on subframes 1 and 2 andit is transmitted every tp seconds. This method is advantageous for the UE,because there is no additional signaling to indicate the location of the ePSS. Thedisadvantage is that it limits the flexibility of resource allocation optimizationat the eNB, because it cannot use the ePSS REs for other data allocation atany scheduling interval.• ePSS on any contiguous band spanning 1.4 MHz: In this variant ofePSS allocation, we use any contiguous band spanning 1.4 MHz within theavailable bandwidth at the eNB for the ePSS. The location of the ePSS canbe broadcast within a System Information Block (SIB), which will be decodedby the UE when it connects to the cell initially and is updated every 3 hours[58]. This is feasible because the ePSS is only used for resynchronization, i.e.,whenever the device wakes up from DRX and reacquires the timing. The initialsynchronization is still done using the legacy synchronization signals. Afterinitial synchronization, the device can decode the SIB, which will convey theePSS location. Another way of conveying the location of the ePSS would bethrough higher layer signaling. Here, the initial location of the ePSS is obtainedfrom SIB, but the subsequent location is indicated to the UE via higher layersignaling. This gives more flexibility for the eNB to optimize the resourceallocation process.3.3.3 ePSS as Quick Sleeping Indication (QSI)Besides the time consumed by the MTC UE for resynchronization, it is highly possiblethat the UE can spend a significant amount of time and power trying to decode thepaging control information on the PDCCH. This is because the LC devices require78Chapter 3. Enhanced Primary Synchronization Signal (ePSS)multiple repetitions to successfully decode the paging control information due to verylow operating SNR.The DRX procedure necessitates the UE to wake-up and look for paging duringeach PO. However, the probability that a UE is paged on every PO instance is verylow since the network consists of a large number of UEs. This problem is alleviated bythe QSI mechanism for energy efficient DRX operation (discussed in Chapter 2). TheQSI mechanism assumes that the UEs are categorized into multiple groups. The QSIsignal will indicate “sleep” if there is no impending page for the UE group, otherwiseit will indicate “stay-awake”. The UE decodes the QSI and attempts to decode thepaging information only if the QSI indicates “stay-awake”. Otherwise, it goes backto sleep immediately and saves power.The earliest phase to indicate the presence or absence of a page is when it is resyn-chronizing with the eNB. Therefore, it would be beneficial if the resynchronizationsignal can also serve as the QSI signal. The ePSS signal presented in the previoussection can fulfill this role as follows.• The LC devices are divided into Ngrp groups and assigned a pair of ePSS pat-terns denoted by (ePSS0, ePSS1).• The eNB schedules the POs one group at a time. It transmits ePSS0 if the UEgroup is not paged in the subsequent PO. Otherwise, it transmits ePSS1.• When the UE attempts to resynchronize with the eNB, it hypothesizes over thetwo assigned patterns and detects one of them. If it detects ePSS0, the UEinterprets that there is no upcoming page in the PO and goes back to sleepimmediately. If it detects ePSS1, then it remains awake to decode the pagingmessage.79Chapter 3. Enhanced Primary Synchronization Signal (ePSS)ePSS not used as QSIePSS used as QSIONNoPagetePSSw tt SPONPagetPDSCHPDCCHtt PDSCHPDCCHtt PDCCHtPSSw tt ePSSw tt Legacy operationPSSw tt DRXtSPNo Page for UE Valid Page for UEPDCCHtONNoDataPONDataPONNoDataPSPPDSCHPDCCHtt ONNoDataPONDataPONDataPFigure 3.5: Illustration of ON time and sleep time for UEs using the legacy and theePSS based resynchronization mechanisms.Hence, the ePSS pattern can also serve as the QSI enabling the UE to resynchronizeand go back to sleep quickly when there is no valid page.Figure 3.5 depicts the UE sleep and wake-up durations for the legacy mode ofoperation, when it uses the ePSS only for resynchronization (not as QSI) and whenit uses the ePSS also as the QSI. The notations tDRX, tPDCCH and tONNoPage denote thelength of UE DRX cycle, the ON time of the UE to decode the PDCCH and the totalON time of the UE when there is no page respectively. Since our ePSS mechanismis designed to provide faster timing reacquisition, the ON time of the UE decreaseswhen it uses the ePSS for resynchronization instead of the legacy synchronizationsignal. Also, the UE ON time further reduces by a factor of tPDCCHtONNoPagewhen the ePSS isalso used as the QSI. A detailed discussion on the performance analysis of the ePSSis presented in the next section.3.4 Performance AnalysisIn this section, we analyze the detection performance of our ePSS mechanism anddemonstrate that it is better than the legacy synchronization signal detection mech-anism. We also provide a simple model based on the ON time to compute the energy80Chapter 3. Enhanced Primary Synchronization Signal (ePSS)consumption of the UE and show that the UEs adopting the DRX mechanism alongwith our ePSS for timing resynchronization consume much lesser energy than theUEs using the current DRX mechanism with legacy synchronization signal detectionfor resynchronization.3.4.1 Reacquisition Performance AnalysisThe ePSS solutions discussed in Section 3.3 were simulated with the settings listedin Table 3.1 and a CFO of 1kHz. The SNR considered for this simulation is -14.2 dB,which corresponds to the operating SNR for the MTC UEs that require 15 dB cover-age enhancement [43]. The detection threshold is set such that the false alarm rateis limited to 1%. Figure 3.6 shows the performance of the legacy synchronizationsignal detection scheme and our two ePSS design schemes (refer to Sections along with the case where the ePSS is also used as the QSI (refer toSection 3.3.3). It is observed that the ePSS designed using longer length ZC se-quences performs slightly better than the ePSS designed by re-using the PSS ZCsequences. As discussed in Section, the longer length ZC sequence has bettercross-correlation properties that results in better performance.In order to analyze the performance, we consider the accumulation time requiredfor 90% detection denoted by tacc. We note that the legacy synchronization signaldetection requires tacc = 420 ms that corresponds to 84 legacy synchronization signalrepetitions, since it is transmitted every 5 ms. In this simulation, the ePSS is trans-mitted on subframe 1 and subframe 2 of the radio frame. This set of two subframesis denoted as “ePSS Block”. For the ePSS constructed using multiple PSS (refer toSection, tacc = 30 ms that corresponds to 3 ePSS Blocks and for the ePSSconstructed using longer length ZC sequences (refer to Section, tacc = 20 ms81Chapter 3. Enhanced Primary Synchronization Signal (ePSS)0 50 100 150 200 250 300 350 400 450 500Accumulation Time(ms) of DetectionEPSS Using Longer ZC (1 pattern)EPSS as QSI Using Longer ZC (2 patterns)EPSS as QSI Using PSS (2 patterns)EPSS Using PSS (1 pattern)Legacy PSS/SSSFigure 3.6: Performance of legacy synchronization signal detection and ePSS detec-tion.corresponding to 2 ePSS blocks. Also, it is evident that there is negligible degrada-tion in performance when the ePSS is also used as the QSI for both the mechanisms.The ePSS detection consumes less than 10% of the time taken by legacy synchro-nization signal detection for the LC devices, which decreases the ON time of the UEsignificantly and reduces the energy consumption.3.4.2 Energy Efficiency AnalysisNow we analyze the energy consumption of the UE following the DRX mechanismwhen it uses the legacy synchronization signal detection for resynchronization andcompare it to the energy consumption of the UE when it uses our ePSS solutions.In this work, we look at energy consumption from the physical layer perspectiveand consider a simple model for our energy consumption calculation based on twoquantities - a) ON time of the UE and b) sleep time of the UE similar to [89].82Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Let tw and tr denote the time taken for the UE RF warm-up and the time taken forthe UE resynchronization respectively. The UE ON time for the PDCCH decode andthe ON time of the PDSCH decode are denoted by tPDCCH and tPDSCH respectively.PON and PS represent the power consumed by the UE during the ON time and thesleep time respectively. In order to decode a page, the UE warms up, resynchronizeswith eNB and decodes the PDCCH to check for a page. If the PDCCH containsP-RNTI, the UE attempts to decode the paging data on the PDSCH. Otherwise,it goes back to sleep. Hence, the total ON time of the UE depends on whether itreceived a page or not. When the UE has a valid page, the ON time (see Figure 3.5)is given by tONPage = tw + tr + tPDCCH + tPDSCH and the corresponding energy consumedis given byEONPage = (tw + tr)PONNoData + (tPDCCH + tPDSCH)PONData, (3.1)where PONNoData and PONData denote the ON time power consumption of the UE withoutactive data and with active data respectively. The warm-up and resynchroniza-tion phases are considered to be non-active data phases since the UE is not run-ning intricate control/data decoding processes like layer demapping, de-precoding.Viterbi/turbo decoding, etc. Therefore, the power consumption in these phases isPONNoData.When there is no page for the UE, the ON time (see Figure 3.5) is determined bytONNoPage = tw + tr + tPDCCH and the corresponding energy consumed is given byEONNoPage = (tw + tr)PONNoData + tPDCCHPONData . (3.2)The sleep time can be obtained by subtracting the ON time from the total lengthof the UE DRX cycle (denoted by tDRX). For the UE using the legacy synchronization83Chapter 3. Enhanced Primary Synchronization Signal (ePSS)signal detection, tr = tPSS, which is the time consumed for the legacy synchronizationsignal detection. For the UE using our ePSS for resynchronization, tr = tePSS, whichcorresponds to the ePSS detection time. Also, when the ePSS is used as the QSI andwhen there is no page for the UE, the decoded ePSS pattern itself suggests that thereis no valid page and the UE does not decode the PDCCH. In this case, the ON timewill be tONNoPage = tw + tePSS. Using p to denote the probability that the UE is paged,the total energy consumed by the UE can be calculated asEtot = p · (EONPage + (tDRX − tONPage)PS)+ (1− p) · (EONNoPage + (tDRX − tONNoPage)PS), (3.3)where PS denotes the power consumed by the UE in deep sleep state.For the Rx power consumption, we used PONData = 500 mW, PONNoData = 250 mW andPS = 0.0185 mW [84, 118]. The UE warm-up time tw was assumed to be 1 ms andthe paging rate p = 0.1. We used tPSS = 420 ms for the legacy synchronization signaldetection, tePSS = 30 ms for the ePSS using multiple PSS and tePSS = 20 ms for theePSS using longer ZC sequences corresponding to the accumulation time required for90% detection (see Figure 3.6). The energy efficiency gain indicates the ratio of theenergy consumed by the UE receiver using the legacy synchronization signal detectionfor resynchronization to the energy consumed by the UE receiver using our ePSS forresynchronization. Table 3.2 summarizes the energy efficiency gain of the UE usingour ePSS solutions for different values of the DRX cycle length tDRX. We examinedtwo scenarios - a) Short paging decode time assuming tPDCCH = tPDSCH = 10 mscorresponding to the upper bound of the energy efficiency gains in Table 3.2 and b)Long paging decode time using tPDCCH = tPDSCH = 40 ms corresponding to the lowerbound of the energy efficiency gains in Table 3.2. The cases we examined include the84Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Table 3.2: Rx energy efficiency gain for LC devices in DRX mode using ePSS.DRX ePSS using ePSS usingLength multiple PSS longer ZCtDRX Not as As Not as AsQSI QSI QSI QSI2.56 s 4.3 - 8.3 10.8 - 12.6 4.7 - 10.6 13.7 - 17.65.12 s 4.3 - 8.3 10.8 - 12.5 4.7 - 10.2 13.6 - 17.510.24 s 4.3 - 8.3 10.7 - 12.4 4.6 - 10.1 13.5 - 17.2327.68 s 3.7 - 6.0 7.5 - 7.9 4.0 - 6.9 8.7 - 9.52621.44 s 2.2 - 2.6 2.8 - 2.9 2.3 - 2.7 2.9 - 3.0maximum length of the DRX cycle supported in the current LTE/LTE-A standards(2.56 s) and the maximum extended DRX cycle length (2621.44 s).In the following, we choose the DRX cycle length of 10.24 s to illustrate theinterpretation of the energy efficiency gains in Table 3.2. In this case, the UE adoptingDRX with our ePSS using multiple PSS mechanism for resynchronization is 4.3 times(denoted by 4.3x) more energy efficient than the UE using DRX with the legacysynchronization signal detection mechanism for resynchronization owing to fastertiming reacquisition using the ePSS. Also, the gain obtained from the ePSS usinglonger ZC sequences is higher compared to the ePSS constructed using multiple PSS(4.7x as opposed to 4.3x), since the ePSS using longer ZC sequences requires a lessertime for reacquisition.Moreover, the energy efficiency improves further when the ePSS is used as the QSI.For example, in the case of the 10.24 s DRX cycle, using our ePSS as QSI mechanismsare 10.7x (for the ePSS using multiple PSS) and 13.5x (for the ePSS using longerZC sequences) more energy efficient than the legacy mechanism respectively. Thisdemonstrates a significant increase compared to the corresponding gains obtainedwhen the ePSS is not used as the QSI (4.3x and 4.6x respectively). This is becausethe LC device does not decode the PDCCH when the ePSS pattern indicates that85Chapter 3. Enhanced Primary Synchronization Signal (ePSS)there is no page (which happens 90% of the time since p = 0.1) unlike the non-QSIcases where the UE has to decode the PDCCH regardless of whether it is paged ornot, thereby simplifying the operation of checking for a page when our ePSS as QSImechanism is used.A common trend observed with the different ePSS solutions is that the energyefficiency gain decreases with the increase in length of the DRX cycle. This is becausethe reduction in energy consumption is directly proportional to the fraction of thetime the UE is ON during the DRX cycle. Since the ON time of the UE depends onthe SNR and not on tDRX, this ratio decreases with increase in tDRX for a fixed SNR.However, even for the maximum DRX cycle length of 2621.44 s, the energy efficiencygain obtained is between 2.2x and 2.7x when ePSS is not used as QSI, which isconsiderable and improves further (between 2.8x and 3.0x) when ePSS is used asQSI. Also, for longer DRX cycle lengths, the legacy UE may have to reacquire theSFN before attempting to decode the paging control information (as discussed later inSection 3.4.5), while the UE using our ePSS solutions (as QSI) is informed about anupcoming page during the timing resynchronization phase itself and need not decodethe SFN when there is no page. The energy efficiency gain computation does notaccount for the ON time of the UE for SFN decoding. Therefore, the energy efficiencygains obtained using our ePSS as QSI solutions (columns 2 and 4 of Table 3.2) forlonger DRX cycles serve as a loose lower bound for the actual gains.3.4.3 Battery Lifetime ImprovementHaving established that the ePSS mechanisms result in higher energy efficiency, wenow estimate the improvement in the battery lifetime of the device when our differentePSS solutions are used. The UE consumes its battery for both transmission and86Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Table 3.3: Battery lifetime gain for LC devices using ePSS with tDRX = 10.24 s andtDRX = 2621.44 s.Tx:Rx ePSS using ePSS usingenergy multiple PSS longer ZCconsumption Not as As Not as Asshare QSI QSI QSI QSItDRX = 10.24 s70:30 1.30 - 1.36 1.37 - 1.38 1.31 - 1.37 1.38 - 1.3950:50 1.62 - 1.78 1.83 - 1.85 1.64 -1.82 1.86 - 1.8930:70 2.16 - 2.60 2.74 - 2.81 2.21 - 2.71 2.84 - 2.94tDRX = 2621.44 s70:30 1.20 - 1.23 1.23 - 1.24 1.20 - 1.23 1.24 - 1.2550:50 1.38 - 1.44 1.47 - 1.49 1.39 - 1.46 1.49 - 1.5030:70 1.62 - 1.76 1.82 - 1.85 1.65 - 1.79 1.85 - 1.88reception. The battery consumption share between UE’s transmission and receptionvaries depending on the UE’s application. For example, a device used for locationtracking has to transmit the data more frequently than that used for smart meterdata reporting. In both these applications require the device to periodically listen tothe network to remain connected. The battery consumption share for transmission ishigher for the UE used for location tracking, since it transmits data more frequently.However, for the UE used for smart meter data reporting, the battery consumptionshare is dominated by periodic reception. Hence, the battery consumption sharebetween UE’s transmission and reception is dictated by the duration of transmissionand reception. Alternatively, if these durations are equal, one module can consumemore energy than the other. For instance, a model where the device transmissionmay consume 2 times more energy than device reception is commonly used. Inthe following, we analyze the improvement in battery life for three different scenariosbased on ratio of the UE battery consumption for transmission to that of the reception- a) 70:30 (Tx > Rx), b) 50:50 (Tx = Rx) and c) 30:70 (Tx < Rx).Since our ePSS mechanisms are adopted by the UEs in the downlink, the energy87Chapter 3. Enhanced Primary Synchronization Signal (ePSS)efficiency gains obtained using our solutions is applicable to the UE reception. Thebattery lifetime gain is calculated asβ =1sTx +sRxη(3.4)where sTx and sRx indicate the UE battery consumption share for transmission andreception respectively, and η is the gain obtained by the use of our different ePSSsolutions (given in Table 3.2). For example, in the first scenario sTx = 0.7, sRx = 0.3and η = 12.4 for tDRX = 10.24 s and ePSS using longer ZC sequences (as QSI).Table 3.3 gives the battery lifetime gain (β) for our different ePSS solutions withtDRX = 10.24 s and tDRX = 2261.44 s in the aforementioned scenarios. In the firstscenario, a conventional battery lasting for 10 years can potentially last 10β = 10 ×1.38 = 13.8 years for tDRX = 10.24 s and 10 × 1.24 = 12.4 years for tDRX = 2621.44 s,when our ePSS using longer ZC sequences (as QSI) solution is adopted. This is asubstantial improvement considering that only 30% of the battery was being used forUE reception. The battery lifetime gain increases as the battery consumption sharefor UE reception increases. For instance, with our ePSS using longer ZC sequences(as QSI) solution in the second and third scenarios, the battery lifetime increasesto 18.6 years and 28.4 years respectively for tDRX = 10.24 s and to 14.9 years and18.5 years respectively for tDRX = 2261.44 s, which is highly significant. The lowestenergy efficiency gain is obtained when our ePSS using multiple PSS (not as QSI)solution is used. Even in this case the lifetime of a conventional battery lasting 10years is extended to 13.0 years, 16.2 years and 21.6 years for the three scenariosfor tDRX = 10.24 s and to 12.0 years, 13.8 years and 16.2 years respectively fortDRX = 2261.44 s, which is considerable. Similar to Section 3.4.2, the battery lifetimeextension for ePSS as QSI solutions for longer DRX cycles serve as a loose lower88Chapter 3. Enhanced Primary Synchronization Signal (ePSS)mePSS Radio FrameRegular Radio Frame(tp/10) - m(a) Frame structure with ePSS transmission ePSS ePSS CE MTC PO  Legacy Paging Opportunities (PO)  10ms Frame CE MTC PO  (b) ePSS and PO AllocationFigure 3.7: Frame Structure and PO Allocation for LC devices using ePSS.bound since the SFN decoding gain is not included in the computation.3.4.4 Analysis of ePSS Transmission OverheadThe periodicity of the ePSS transmission determines the overhead on the network.If the ePSS is scheduled on every radio frame like the PSS, it will consume a lotof network resources. From our simulation results (see Figure 3.6), we note that mePSS frames are required for the desired accuracy of timing detection, where m = 3for ePSS using PSS and m = 2 for ePSS using longer ZC sequences. If the periodicity89Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Table 3.4: Network resource overhead due to the ePSS transmission.eNB Bandwidth tp = 100 ms tp= 360 ms1.4 MHz (6 PRB pairs) [4% 6%] [1.11% 1.67%]5 MHz (25 PRB pairs) [0.96% 1.44%] [0.26% 0.4%]10 MHz (50 PRB pairs) [0.48% 0.72%] [0.13% 0.2%]20 MHz (100 PRB pairs) [0.24% 0.36%] [0.06% 0.1%]of ePSS is tp ms, the transmission consists m ePSS radio frames followed by (tp10−m)legacy radio frames as shown in Figure 3.7a.The ePSS is mostly used by the LC devices. Therefore, to minimize the resourceconsumption, we propose to group the POs of such UEs close to the ePSS. Thissolution is illustrated in Figure 3.7b. For a given interval of time, the number of LCdevices might a small subset of the total number of UEs. Assuming that 10% of theUEs are LC devices, the ePSS only needs to be sent every tp = 100 ms. Also, theP-RNTI for the LC devices can be sent on the enhanced PDCCH (ePDCCH) [117],which uses the subcarriers in the PDSCH space. The POs for the legacy UEs will bedistributed and they will be paged via the regular PDCCH (PDCCH). This way thePDCCH capacity is not altered.Moreover, when the ePSS is not transmitted every radio frame, a single ePSS willbe used to resynchronize UEs that have their PO subframes greater than 10 ms fromthe ePSS subframe. Therefore, the UE can save power by going back to sleep aftersuccessfully detecting the ePSS and wake up again just before its PO. However, wehave to ensure that the UE does not lose the symbol timing (UE clock does not driftby more than 5% of the symbol time, which is 3.6 µs) during its sleep time. Forexample, for a UE using a crystal with 10 ppm accuracy, the clock drift is within 5%of a symbol if the sleep time is less than 360 ms. Therefore, the maximum value oftp is 360 ms for a 10 ppm accurate UE clock.The ePSS consumes 6 PRB pairs × 2 subframes = 12 PRB pairs every 10 ms (refer90Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Figure 3.4a). As seen from Figure 3.6, the time taken to decode ePSS can be 20 ms(= 24 PRB pairs for ePSS using longer ZC sequences) or 30 ms (= 36 PRB pairs forePSS using multiple PSS). The overhead on the eNB due to the resource allocation forthe ePSS with periodicity tp = 100 ms and tp = 360 ms is summarized in Table 3.4.The numbers in each row of the second and third columns of Table 3.4 correspondto the network load considering ePSS decode time of 20 ms and 30 ms respectively.The eNB deployments usually have a bandwidth of 5 MHz or more and the networkoverhead due to the ePSS for such eNBs is less than 1.5% for both tp = 100 msand tp = 360 ms. Moreover, the overhead of ePSS transmission is restricted to theidle mode. The regular data transmission occurs after the UE transitions from theidle mode to the connected mode, where the resource allocation uses the defaultLTE/LTE-A frame structure, without the ePSS. Therefore, the ePSS does not affectthe regular data transmission.3.4.5 DiscussionIn summary, our ePSS solutions reduced the energy consumption when compared tothe legacy solutions due to the following reasons.• The ePSS consists of an increased density of synchronization signals, i.e., ePSSpacks more synchronization signals in less amount of transmission time, therebyreducing the time taken by the UE to reacquire the symbol timing.• The ePSS has an additional attribute of being able to indicate whether there isan impending page for the UE or not, thereby helping UEs to return to sleepmode quickly when there is no page and save energy. But the legacy UE followsthe procedure of decoding the P-RNTI on the PDCCH regardless of whether it ispaged or not, which is complex and energy consuming.91Chapter 3. Enhanced Primary Synchronization Signal (ePSS)• The ePSS with best performance uses longer length ZC sequences, which possessbetter signal detection properties, thereby improving the reacquisition time andreducing power consumption.Moreover, the legacy UE may have to reacquire the SFN so that it attemptsto decode the paging information on the correct frame/subframe, depending on theaccuracy of its clock and the length of the DRX cycle. This is because the UE willlose its frame timing when its clock drifts by more than half a frame (= 10ms2= 5 ms).For example, an LC UE with 10 ppm accurate clock [116] will lose its frame timingif it sleeps for more than 5ms(10·10−6) = 500 s. When the DRX cycle length is than 500 s,the clock drift results in a symbol timing offset, but the UE remains in the same radioframe. However when the DRX cycle is longer than 500 s, both the frame timing andthe symbol timing are lost and the UE has to reacquire the SFN.The SFN field in LTE/LTE-A consists of 10 bits and is incremented by one every10 ms (= frame length) [58, 59]. Hence the maximum time length that can be indi-cated by the SFN field is 210 × 10−3 = 10.24 s. In order to support extended DRXcycles, the on-going standardization activities in LTE/LTE-A have included a newfield called the Hyper - System Frame Number (H-SFN) which incremented by oneevery 10.24 s [56]. Currently, the number of H-SFN bits is set to 8. Consequently,the maximum time length that can be indicated with the combination of H-SFN andthe legacy SFN is (28 − 1) × 10.24 + 10.24 = 2621.44 s, which corresponds to themaximum extended DRX cycle length.The UE has to decode the H-SFN and the SFN if it sleeps so long that theframe timing is lost, and then attempt the paging decode on the correct pagingframe/subframe. Owing to the low operating SNR, H-SFN/SFN decoding can takemultiple repetitions, increase the ON time and the energy consumption of the UE.92Chapter 3. Enhanced Primary Synchronization Signal (ePSS)Also, the legacy UE needs to perform this decoding even when there is no pagebecause it can know about the validity of the page only when it decodes the pagingcontrol information (which is post H-SFN/SFN decoding). But, when our ePSS asQSI solutions are used, the UE is informed about the validity of the page duringthe timing resynchronization phase itself and H-SFN/SFN decoding would not benecessary when there is no page. The savings obtained from not decoding the H-SFN/SFN when there is no page has not been included in our calculations for theenergy efficiency gain and battery lifetime extension. Therefore, our results for thiscase serve as a loose lower bound and the actual gains that can be obtained usingour solutions can be higher when the DRX cycle length is longer (>500 s for an LCUE with 10 ppm accurate clock).Furthermore, our ideas can be extended to other OFDM based systems wherethe conventional methods of CP correlation and reference signal detection take along time to converge at very low SNR. For example, in WLAN, two new, closelyspaced, robust preamble sequences can be assigned to each UE, one indicating a“Page” and the other indicating “No Page”. This mapping of paging indication toresynchronization sequences, combined with shrewd resource allocation would resultin an energy efficient operation of the MTC UEs.3.5 ConclusionIn this chapter, we considered the DRX mechanism in LTE/LTE-A for the low-cost,low-complexity MTC UEs requiring coverage enhancement (denoted by LC devices).We explored the sensitivity of the paging decode operation to the timing offset ingreater detail by improving the timing resolution from being symbol accurate (seeChapter 2) to being sample accurate. Firstly, we showed that the paging decode93Chapter 3. Enhanced Primary Synchronization Signal (ePSS)operation, which is a critical part of the DRX mechanism, is very sensitive to timingoffset. Secondly, we explored the conventional timing acquisition algorithms - CPautocorrelation and the legacy synchronization signal detection and demonstratedthey take around 600 ms and 380 ms respectively for reacquiring the symbol timingwithin tolerable limits. This leads to increased ON time of the UE and hence re-sults in increased energy consumption. To mitigate this problem, we introduced theePSS within the LTE/LTE-A standardization framework and proposed a novel DRXmechanism which uses our ePSS for faster timing resynchronization and reduced en-ergy consumption. We described two simple methods to design the ePSS signal -by using multiple PSS sequences and by using ZC sequences and indicated that ourePSS design requires less than 1.5% network resource overhead for eNB bandwidthsof 5 MHz or more.Further, adopting the DRX with QSI mechanism described in Chapter 2, weproposed the use of the ePSS as QSI to further improve the energy efficiency of theLC devices. We illustrated that the modified DRX mechanism for the LC devicesusing our ePSS solutions results in 1.2 to 1.8 times improvement in their batterylifetime. Similar to the QSI mechanisms from Chapter 2, the specific ePSS solutionsare suitable for UE categories of CAT-M1, CAT-0 and above. But the core ideabehind the ePSS, i.e., faster resynchronization and quick return to sleep, is applicableto all the UE categories. For example, a synchronization signal block occupying 12subcarriers × 11 symbols is being considered as the PSS in NB-IoT standardization,in order to reduce the time consumed for initial acquisition. Also, studies in 5G NRare accounting for the energy consumption due to timing acquisition and consideringnovel designs for the synchronization and paging channels.94Chapter 4Low SNR Uplink CFO Estimation4.1 IntroductionIn Chapter 2 and Chapter 3, we described energy efficient mechanisms for the LTE/LTE-A MTC UEs in the downlink. In this chapter, we explore the MTC mechanisms forenergy efficient uplink transmission using LTE/LTE-A, under low network coverage.Although the UEs belonging to CAT-M1 or higher categories could operate on abandwidth of at least 1.4 MHz, the LTE/LTE-A MTC standardization activities inthe uplink focused on single PRB transmission (occupying 180 kHz bandwidth), inorder to address massive access by supporting higher number of MTC UEs per timeinterval. The introduction of the NB-IoT framework (described in Section 1.2.2) tothe LTE/LTE-A standards provided the further motivation to device efficient UEdata transmission mechanisms over a narrow bandwidth. Therefore, single PRBtransmission became a common theme across different UE categories, promoting theexchange of ideas between standardization activities for CAT-0, CAT-M1 and NB-IoT UEs. Moreover, developing mechanisms to reduce the number of retransmissionsand hence improve the uplink energy efficiency became a common challenge to beaddressed in all UE categories.It is well known that improved channel estimation, including accurate timingand frequency offset estimation is beneficial for data decoding. As mentioned inSection 1.2.2, the eNB adopts the timing advance indication mechanism to ensure95Chapter 4. Low SNR Uplink CFO Estimationthat all UE transmissions are time synchronized. Minor deviations in timing offset areestimated accurately using the CP autocorrelation [63,64]. These procedures ensurethat the residual timing offset is within the CP length [58] and do not impact thesystem performance. However, the tracking the Carrier Frequency Offset (CFO) ofeach UE using CP autocorrelation is complex (as described in detail in Section 4.3.1),necessitating novel mechanism for CFO estimation at the eNB.Prior works for uplink CFO estimation consider normalized CFO (actual CFOvalue divided by the subcarrier spacing) and describe algorithms to estimate andcompensate for the normalized CFO [63, 64, 107, 119]. However, there is a smallportion of the CFO remaining in the system after compensation depending on theaccuracy of estimation, called the residual CFO. Although it is desirable to minimizethis residual CFO, the benefits from this reduction are minimal at high operatingSNR, since the channel estimation algorithms at the base-station can effectively han-dle a small residual CFO. In LTE/LTE-A, the subcarrier spacing is 15 kHz [58]. Forinstance, when the actual residual CFO in the system is 100 Hz, the normalizedresidual CFO value is 0.067, which is negligible at high SNR. However, a significantnumber of MTC UEs are expected to operate in low coverage areas, where the op-erating SNR is as low as -15 dB [43, 44] and as we demonstrate, the residual CFOadversely impacts the performance of such UEs.In this chapter, we first demonstrate that further reduction in residual CFO ishighly beneficial for MTC UEs in low coverage, since it reduces the number of dataretransmissions, which in turn reduces the ON time of the UEs and hence increasesthe energy efficiency. We propose ML based algorithms for robust CFO estimation inlow coverage, using - a) repeated data transmission, b) pilot transmission and c) bothrepeated data and pilot transmission. We also derive the Crame´r-Rao lower bound96Chapter 4. Low SNR Uplink CFO Estimationfor these estimators. Next, we provide an insight into the NB-IoT large transportblock transmission scheme, which is a novel, simple and effective mechanism beingconsidered by the 3GPP for energy efficient MTC and illustrate how the energyefficiency obtained from this scheme can be further enhanced by the use of our robustCFO estimation techniques. Lastly, we propose a variation of the LTE/LTE-A framestructure incorporating additional pilot signals during the initial MTC transmissions,which assists in faster CFO estimation at the eNB with minimal overhead.The rest of the chapter is organized as follows. In Section 4.2, we analyze the effectof CFO on the energy efficiency of the NB-IoT MTC UEs. In Section 4.3, we describethe conventional techniques used for CFO estimation and in Section 4.4, we introduceour ML based CFO estimation technique for LTE/LTE-A MTC. Using simulations,we compare the performance of our CFO estimation technique with the conventionaltechniques in Section 4.5, followed by a detailed analysis of the energy efficiencyof MTC UEs using CFO estimation and compensation for the current transmissionscheme and the newly proposed large transport block transmission mechanism. InSection 4.6, we propose a new MTC transmission technique with increased pilot den-sity, which uses our ML based CFO estimation technique for faster CFO estimationin low coverage. The conclusions are presented in Section Effect of CFO Estimation for NB-IoT UplinkAs described in Section 1.2.2, the NB-IoT mechanism in LTE/LTE-A refers to thetransmission and reception of signals on a bandwidth corresponding to one PRB(180 kHz). This is helpful to low data rate MTC UEs, where the transmission happensin small bursts of data followed by long idle periods. It would be ideal for the UEto complete its data transmission with minimal number of retransmissions, switch97Chapter 4. Low SNR Uplink CFO Estimationto idle mode and save power. The number of retransmissions depends on the SNR,the channel conditions and the timing/frequency estimation accuracy at the eNBdecoding the UE data. While the CP autocorrelation and timing advance mechanismsenable the eNB to accurately track the timing offset [63,64], tracking the CFO usingCP autocorrelation would be quite complex (explained later in Section 4.3.1).In this section, we demonstrate our model for energy consumption analysis anddetermine the effect of residual CFO on the energy consumption using numericalcalculations.4.2.1 Energy Consumption ModelIn order to calculate the energy consumption, we adopt a simple model,E = PONtON + POFFtOFF (4.1)where PON and POFF denote the power consumed by the UE during its active (ON)period and sleep (OFF) period respectively. The durations ON and OFF periods arerepresented by tON and tOFF respectively. The total time length is tTotal = (tON+tOFF)and we define v = POFFPON, where v  1 since the sleep time power consumption ismuch lower than the active time power consumption. Then, the energy consumptionis calculated asE = PON(tON + v(tTotal − tON)). (4.2)4.2.2 Numerical ResultsTo illustrate the impact of CFO on the UE energy consumption, we consider a scenariowith no residual timing offset, since it can be estimated with a sufficient degree of98Chapter 4. Low SNR Uplink CFO Estimationaccuracy using the CP and determine the number of repetitions required for an MTCUE for different values of the residual CFO. The presence of CFO leads to InterCarrier Interference (ICI), which affects the performance. However, the amount ofICI is small, since we analyze the effects of residual CFO. Multi-user Interference(MUI) is not considered because the LTE/LTE-A standards ensure that subcarriersare not shared between different users for a given subframe [58]. We first analyzethe performance for CFO = 100 Hz, which is the value used for MTC performanceevaluation by the 3GPP [43] and then for lower values of CFO, corresponding to50 Hz, 25 Hz and 10 Hz. Among the different CFO values, we use CFO = 10 Hzto model the scenario where the frequency offset is negligible, based on simulationswhich indicated that the number of repetitions required by the UE did not changesignificantly for 0 ≤ CFO ≤ 10 Hz.The simulations are performed using the LTE toolbox in MATLAB. In order toanalyze the low coverage scenario, the 3GPP recommends the evaluation of perfor-mance for 18 dB additional coverage [43], which corresponds to our operating SNRof -15.5 dB. For the channel model, we use the EPA model with a Doppler spreadof 1 Hz, which is advocated by the 3GPP for MTC UEs with limited mobility [43].We use a single PRB pair transmission scheme (12 subcarriers × 1 subframe) withthe MCS index chosen to be 5 (corresponding to QPSK modulation and a code rateof 0.4385 [57]), consistent with the 3GPP recommendation for NB-IoT. We use TBS= 72 bits, which is the maximum transport block size for MCS = 5 [57]. Othersimulation parameters are summarized in Table 4.1.Table 4.2 gives the number of subframes required by the eNB to decode thetransport block and the effective data rate for different values of the residual CFO. Itis evident that a transport block with lower CFO requires fewer repetitions than the99Chapter 4. Low SNR Uplink CFO EstimationTable 4.1: Simulation parametersParameter ValueNo. data symbols 12 subcarriers ×per subframe 12 symbolsNo. of DMRS symbols 12 subcarriers ×per subframe 2 symbolsNo. UE antennas 1No. eNB antennas 2eNB bandwidth 10 MHzeNB sampling rate 15.36 MHzChannel Model EPA 1 HzSNR -15.5 dBBLER 0.1Table 4.2: Energy efficiency vs. CFO for TBS = 72 bitsCFO NSF Reff (kbps) Energy Efficiency GainD = 10% D = 1% D = 0.1%100 110 0.44 - - -50 100 0.48 8.6% 8.5% 7.5%25 92 0.52 15.5% 15.3% 13.5%10 80 0.60 25.9% 25.5% 22.5%one with higher CFO, thereby increasing the effective data rate. This suggests thata lower residual CFO at the eNB helps the MTC UE to complete its transmissionquickly, turn off its radio and save power.We obtain the energy consumed by the UE for the different CFO values usingEq. (4.2) with tON =1Reffs per bit, where Reff is the effective data rate given by 1.1.We consider the energy consumed by the UE for CFO = 100 Hz, denoted by Eorig,as our reference and compute to the energy efficiency gain as(1− EnewEorig), where Enewis the energy consumption of the UE corresponding to the lower CFO values.Table 4.2 summarizes the energy efficiency results when tTotal is an integer multipleof torig, which is the time taken for successful decoding with CFO = 100 Hz. That is,tTotal = qtorig and q = 10, 100, 1000, which correspond to duty-cycles (D) of 10%, 1%and 0.1% respectively, PON = 100 mW (corresponding to the 20 dBm transmission100Chapter 4. Low SNR Uplink CFO Estimationpower of MTC UEs [43, 48]) and POFF = 0.015 mW (based on the sleep time powerconsumption indicated in [84,118]). The reason to choose these duty-cycles is that foreach case, the inactive duration of the UE is much greater than the active duration,which suitably models the infrequent data transmission and low-data rate mode ofoperation of MTC UEs.A common trend that we note in these results is that the energy efficiency gaindecreases with decrease in D. This is intuitive because the reduction in energy con-sumption is obtained by reducing the ON time of the UE and smaller values of Dresults in lower ON time. We observe that the energy efficiency gain increases with adecrease in residual CFO. The reduction of residual CFO from 100 Hz to 10 Hz resultsin 22.5% reduction in energy consumption even for a low duty cycle of 0.1%, which issignificant. Therefore, a robust CFO estimation mechanism at the eNB, which worksaccurately at low operating SNRs and helps in reducing the energy consumption ofthe MTC UE is desirable.4.3 Conventional CFO Estimation TechniquesHaving established the need for accurate CFO estimation to enable high energy ef-ficiency of IoT communication, we now discuss the CFO estimation techniques thatare currently used in the uplink. In particular, we consider two techniques - a) CPautocorrelation [63, 64] and b) symbol repetition demonstrated in [107, 119] and thereferences within, which are widely used for fractional frequency offset estimation inthe uplink. We illustrate why these techniques cannot be used by MTC UEs usingLTE/LTE-A in low coverage.In literature, fractional frequency offset is often represented and estimated interms of the normalized CFO, i.e., the actual CFO value divided by the subcarrier101Chapter 4. Low SNR Uplink CFO Estimationspacing (∆F ). The subcarrier spacing is related to the sampling rate (Ns) and theFFT size (NFFT) such that Ns = NFFT∆F . However, we choose to represent thefrequency offset using actual CFO instead of normalized CFO because our workconsiders the estimation of residual CFO, which is typically represented in terms ofthe actual value.4.3.1 CP AutocorrelationIn OFDM based systems, the CFO is estimated from the phase of the autocorrelationof the CP asˆ =Ns2piNFFTangleNCP−1∑n=0y(n+NFFT)y∗(n) (4.3)where y(n) is the nth sample of the received time-domain signal at the eNB, Nsis the sampling rate, NFFT is the FFT size used at the eNB and NCP is the CPlength [63, 64]. From Eq. (4.3), we see that ˆ is the product of the normalized CFOwith the subcarrier spacing (indicated by the NsNFFTscaling factor), which denotes theactual CFO in the system.In multiple access systems like Orthogonal Frequency Division Multiple Access(OFDMA) and Single Carrier - Frequency Division Multiple Access (SC-FDMA),when multiple UEs occupy the spectrum, the time-domain symbol and the CP con-tains components from all the UEs . Assuming that UEs have perfect timing syn-chronization, the CP portion of the received signal at the eNB will consist of the sumof the CPs of all the UEs. Each UE might have a different CFO. Therefore, detectionof each UE’s CFO requires the separation of its time-domain symbol and its CP fromthe multiplexed received signal.In order to get the per-UE time-domain symbol, the eNB first takes an FFT of the102Chapter 4. Low SNR Uplink CFO EstimationSymbol s1CP c1Symbol s2CP c2UE 1UE 2UE KCFO = f1CFO = f2CFO = fKSymbol sKCP cKΣ Symbol sCP cIFFTIFFTIFFTFrequency Domain Time DomainMultiplexed time domain signalFFTInsert zerosRetainInsert zerosRetainRetainInsert zerosSeparate  frequency components of each UEIFFTIFFTIFFTSymbol s1CP c1Symbol s2CP c2Symbol sKCP cKEstimate f1Estimate f2Estimate fKCorrelateFrequency DomainFor each UE, subtract the sum of the signals from all the other UEs from the received, multiplexed time domain signal to get the individual time domain signal.Figure 4.1: Illustration of CFO estimation using CP autocorrelation.multiplexed time-domain signal, retains the subcarriers of the UE of interest, sets theremaining subcarriers to zero and takes an Inverse Fast Fourier Transform (IFFT).This procedure is illustrated in Figure 4.1. Moreover, in the case of MTC UEs inlow coverage, multiple repetitions of the time domain symbol and CP are requiredfor successful detection, which further increases the complexity. For example, aneNB with a bandwidth of 10 MHz has 50 PRBs available for user data and uses a1024-point FFT. For MTC UEs using single PRB transmission and a large number ofMTC devices present in the network, we can potentially have 50 UEs served at eachinstant. To separate the time-domain symbol and CP of each UE, the eNB requires 1FFT and 50 IFFTs. Since the FFT/IFFT is O(N log2(N)) complex operations, thisrequires 51 × 1024 × 10 ≈ 5.2 × 105 complex operations, which is computationallyintensive. Furthermore, if the UEs are in low coverage and assuming that 14 symbols(1 subframe) are required for successful frequency offset detection, the number ofcomplex operations increases to 7.3 × 106. Therefore, CP autocorrelation is not anideal candidate for CFO estimation in the case of MTC UEs.4.3.2 Symbol RepetitionBesides the CP autocorrelation method, CFO estimation can be done by correlatingrepetitions of data or pilot signals and measuring the correlation phase angle (see103Chapter 4. Low SNR Uplink CFO EstimationReference SymbolRepetition periodCombine consecutive repetitionsSub-carriersFigure 4.2: Illustration of CFO estimation using symbol repetition.Figure 4.2) [107, 119]. However, the repetitions should be close enough in time, sothat the phase angle does not roll-over. If a UE has to measure a CFO ranging from-f0 Hz to f0 Hz, the maximum amount of time between two repetitions is given byTrep =12f0. For example, we require Trep = 1 ms, if the UE has to measure a CFOranging from -500 Hz to 500 Hz. In other words, the detectable CFO range decreasesas Trep increases. The estimation method is formulated asˆ =Ns2piNg(angle(N−1∑n=1Yn · Y ∗n−1))(4.4)where Y is the frequency-domain received symbol spanning over Nsc subcarriers, “·”denotes element-wise multiplication, n indicates the repetition index, N is the numberof repetitions required to successfully detect the CFO, Ns is the sampling rate andNg is the number of samples between the consecutive symbol repetitions in termsof the FFT size, NFFT. For example, when the repetitions occur every 2 symbols,Ng = 2NFFT. Again, ˆ in Eq. (4.4) also denotes the actual CFO in the system.Unlike the method of CP autocorrelation, this technique can be scaled to accom-modate multiple UEs. This is because the method uses frequency-domain symbols104Chapter 4. Low SNR Uplink CFO Estimationand the signals of different UEs can be easily separated and the CFO of each UEcan be separately calculated in the frequency domain. However, in LTE/LTE-A, theDMRS symbols repeat every 10 ms and the range of the CFO that could be detectedwith this is only from -50 Hz to 50 Hz, which is smaller than the residual CFO range(-100 Hz to 100 Hz) in the system. Therefore, the DMRS symbols cannot be directlyused for correlation phase angle based CFO estimation. In the following, we pro-pose our mechanisms for CFO estimation for MTC UEs using LTE/LTE-A in lowcoverage.4.4 ML Based CFO estimationIn this section, we describe the design of our ML based CFO estimation algorithmfor two cases - 1) using repeated RV transmission and 2) using the DMRS. Our MLbased CFO estimation method is an extension of the method discussed in [119], whichwas designed for consecutive symbol repetition. We modify the algorithm in [119]so as to fit the LTE/LTE-A frame structure and operate on subframe repetitions (incase 1) and reference signal repetition (in case 2).Let d denote the transmitted signal of length K samples. The CFO of the UE isdenoted by . Since the CFO is a phase-ramp in time-domain, the signal with CFOis given bys(k) = d(k)e(j2pikNs) = d(k)ejkθ (4.5)where Ns is the sampling rate, k = 0, 1, 2, · · · , K − 1 andθ =2piNs. (4.6)Using Eq. (4.5), the LTE/LTE-A transport block transmission in time-domain can105Chapter 4. Low SNR Uplink CFO Estimationbe expressed assn(k) = dn(k)ej(k+nK)θ (4.7)where k = 0, 1, 2, · · · , K − 1, n = 0, 1, 2, · · ·NSF − 1 and NSF is the number ofsubframes required for the successful decoding of the transport block.The current LTE/LTE-A standards support 4 RVs of the UE data block to betransmitted. The UE transmits one RV per subframe and the RV index is cycled inthe order [0, 2, 3, 1], i.e., d0 = d4 = d8 = · · · = r0, d1 = d5 = d9 = · · · = r2 and soon, where rq denotes the RV being transmitted with the RV index q = 0, 2, 3, 1. Thismeans that the RV is repeated every 4 subframes and considering that each subframeis 1 ms, the range of CFO detection is -125 Hz to 125 Hz.In the ongoing LTE MTC standardization, RV repetition is being proposed forMTC UEs. When the UE uses RV repetition, it respects the standard RV cycling or-der, but can transmit N repetitions of the same RV index before switching to the nextindex, i.e, d0 = d1 = d2 = · · · = dN−1 = r0, dN = dN+1 = dN+2 = · · · = d2N−1 = r2and so on. For example, if N = 3, the UE transmits [0,0,0,2,2,2,3,3,3,1,1,1,0,0,0,. . . ].Therefore, for the MTC UEs, the CFO detection range is -500 Hz to 500 Hz.4.4.1 ML Based CFO Estimation Using Repeated DataIn the following, we derive an ML based technique, which uses the RV repetitions toestimate the CFO. We define a new signal x, which consists of N repetitions of thesame RV (denoted by r). Then, we havexn(k) = r(k)ej(k+nLK)θ (4.8)106Chapter 4. Low SNR Uplink CFO Estimationwhere k = 0, 1, 2, · · · , K−1, n = 0, 1, · · ·N−1, L = 4 for legacy UEs (since the sameRV is repeated every 4 subframes) and L = 1 for MTC UEs (since the repetitionsare consecutive).Let R denote the DFT of r(k)ejkθ. Then, in frequency domain, each RV receptionat the eNB can be expressed asYn = Hn ·RejnLKθ +Wn (4.9)where Hn is the channel vector (n = 0, 1, · · · , N − 1), Wn is the noise vector andHn ·R denotes the element-wise multiplication between Hn and R.We assume that the channel remains the same for N subframes, which holds inthe case of pedestrian channels. Therefore, Hn = H,∀n. In order to estimate theCFO, we have to estimate θ from Eq. (4.9). Since we have no information about thedata and the channel, the unbiased estimate for the vector H ·R is given byCˆ =1NN−1∑n=0Yne−jnLKθ. (4.10)Substituting Eq. (4.10) to Eq. (4.9), the ML estimator for the phase angle θ, denotedby θˆ and the corresponding CFO estimate (ˆ) are given byθˆ = minθN−1∑k=0‖Yk − Cˆ · ejkLKθ‖2, (4.11)ˆ =θˆNs2piLK. (4.12)The value of θˆ is obtained by searching over the different values of θ between 0and 2pi in discrete steps. The step size is set according to the required resolution of107Chapter 4. Low SNR Uplink CFO Estimationthe CFO estimate. If the CFO resolution is fr Hz, then the step size for the searchis 2piLKfrNs. The complexity of the search increases when a finer resolution is requiredfor the CFO estimate. Crame´r-Rao Lower BoundThe performance of an estimator is typically analyzed using the Mean Squared Error(MSE), which is lower bounded by the Crame´r-Rao bound. For our ML based CFOestimator using repeated data, Crame´r-Rao bound is given byCRB =3N2sΨ−14pi2L2K2MN(N − 1)(4N − 3) (4.13)where Ψ is the SNR and M is the number of DFT samples used for estimating theCFO. The procedure to derive CRB() is illustrated in the Appendix A.4.4.2 ML Based CFO Estimation Using the DMRSThe generic structure of our ML based CFO estimation technique enables us toextend its applicability to the periodic repetitions of DMRS signals in LTE/LTE-A. A DMRS symbol is transmitted every half subframe. For DMRS transmission,Eq. (4.9) changes toY˜nm = Gnm · Pnmej(2n+m)Kθ2 + W˜nm (4.14)where Pnm are the known DMRS sequences, Gnm and W˜nm are the channel and thenoise vectors with n = 0, 1, · · · , N − 1, denoting the subframe index and m = 0, 1indicates whether the DMRS is transmitted on the first half (m = 0) or the secondhalf of the subframe (m = 1). Therefore, L is set to 12for DMRS transmission and108Chapter 4. Low SNR Uplink CFO Estimationthere is no difference between the legacy and MTC UEs. This is because the DMRSis transmitted in the same manner for legacy as well as MTC UEs with a periodicityof half subframe (0.5 ms).Now, we derive the ML based CFO estimator using the DMRS. Similar to the MLestimator for repeated data, we assume that the channel does not vary over the Nsubframes of interest. Hence, Gnm = G,∀n,m. Then, the channel estimate is givenbyGˆ =12NN−1∑n=01∑m=0Y˜nm · P ∗nme−j(2n+m)Kθ2 , (4.15)and the ML estimator for θ is given byθˆ = minθN−1∑k=01∑l=0‖Y˜kl − Gˆ · Pklej(2k+l)Kθ2 ‖2, (4.16)and the corresponding CFO estimate can be calculated using Eq. (4.12). The rangeof CFO values that can be detected using this mechanism is between -1 kHz to 1 kHz,since the DMRS periodicity is 0.5 ms. The Crame´r-Rao bound for this case can alsobe obtained from Eq. (4.13), using L = 12, M equal to the length of the DMRSsequence and 2N repetitions instead of N .4.4.3 Modified Conventional CFO Estimation Scheme forDMRSAlthough DMRS symbols are transmitted every 0.5 ms in LTE/LTE-A, the durationof repetition between identical DMRS symbols is 10 ms. If the conventional correla-tion phase angle method is used on these DMRS repetitions, it results in a reducedCFO detection range of -50 Hz to 50 Hz (refer to Section 4.3.2). Here, we suggesta modification to the conventional method so that it can make use of all the DMRS109Chapter 4. Low SNR Uplink CFO Estimationtransmissions to estimate the CFO within the desired range.We multiply each received DMRS symbol (Y˜nm in Eq. (4.14) by the conjugate ofthe reference DMRS symbol (Pnm) and obtain the CFO estimate by using the phaseangle of the correlation of consecutive DMRS symbols. To illustrate this mechanism,we denote Z2n+m = Y˜nm · P ∗nm, ∀n,m, where n = 0, 1, · · ·N − 1 and m = 0, 1. Then,the CFO estimate is given byˆconv =NspiK(angle(2N−1∑l=1Zl · Z∗l−1)). (4.17)The range of CFO detection using such a modified mechanism is between -1 kHz and1 kHz, similar to the ML based CFO estimation technique using DMRS.4.4.4 ML Based CFO Estimation Using Repeated Datawith DMRS CompensationOur ML based CFO estimation using repeated data proposed in Section 4.4.1 usesonly the data symbols for estimating the CFO. The DMRS symbols are not usedbecause they are not the same between consecutive subframes. Here, we extend thismethod such that it also incorporates the DMRS symbols. This is done by multiplyingeach received DMRS symbol by the conjugate of the reference DMRS symbol (similarto the method in Section 4.4.3). Then, all the DMRS symbols will be a vector ofones, multiplied by the channel co-efficient and the CFO in that symbol plus thenoise at the receiver. This will give us two additional symbols per subframe for MLestimation of CFO.110Chapter 4. Low SNR Uplink CFO Estimation4.5 Simulation ResultsIn this section, we first present the simulation results for our ML based CFO estima-tion algorithms and compare their performance with the conventional CFO estimationtechniques. Then, we introduce the large transport block transmission mechanismfor MTC, where the UE transmits transport blocks whose size is larger than thatsupported in the current LTE/LTE-A standards. We illustrate that this mechanismimproves the effective data rate of the UE and reduces the energy consumption. Wealso show that the energy efficiency of the MTC UE is further enhanced when thismechanism is used in conjunction with our ML based CFO estimation technique.4.5.1 Performance of CFO Estimation TechniquesIn order to analyze the performance of our ML based CFO estimation and the conven-tional CFO estimation techniques, we consider three cases - a) using data symbolsonly, b) using the entire subframe with DMRS compensation and c) using DMRSsymbols only. In the first case, we have 12 symbols available per subframe for CFOestimation and in the second case, the DMRS symbols of the received subframe aremultiplied with their conjugates, so that the entire subframe can be used for CFOestimation. In the third case, we use only the 2 DMRS symbols in each subframe toestimate the CFO. The residual CFO in the system is 100 Hz and the CFO estimationerror is measured as the absolute value of the difference between the actual CFO andthe estimated CFO values. We evaluate the performance based on the number ofsubframes required to estimate the CFO within 10 Hz accuracy, denoted by NCFO.111Chapter 4. Low SNR Uplink CFO EstimationSNR in dB-20 -19 -18 -17 -16 -15 -14 -13 -12 -11 -10MSE10 -310 -210 -110 010 110 210 310 4CRBSimulationN = 16N = 32N = 64N = 128(a) Using data symbols onlySNR in dB-20 -19 -18 -17 -16 -15 -14 -13 -12 -11 -10MSE10 -310 -210 -110 010 110 210 310 4CRBSimulationN = 16N = 32N = 64N = 128(b) Using the entire subframeSNR in dB-20 -19 -18 -17 -16 -15 -14 -13 -12 -11 -10MSE10 -310 -210 -110 010 110 210 310 4CRBSimulationN = 16N = 32N = 64N = 128(c) Using DMRS symbols onlyFigure 4.3: MSE vs. SNR and Crame´r-Rao bound for ML based CFO estimation inAWGN.4.5.2 MSE and Crame´r-Rao Bound for the GaussianChannelFirst, we determine the MSE of our ML based CFO estimator for the three casesin an AWGN channel and compare the MSE with the Crame´r-Rao bound given inEq. (4.13). The eNB bandwidth is chosen to be 10 MHz and the correspondingsampling rate Ns = 15.36 MHz. Therefore, each subframe (1 ms) contains K = 1 ms× 15.36 MHz = 15360 samples. In the first two cases, the data is repeated in everysubframe (MTC RV transmission case), which corresponds to L = 1. For the first112Chapter 4. Low SNR Uplink CFO Estimationcase, the number of DFT samples used for CFO estimation, we have M = 12 symbols× 12 subcarriers = 144 and for the second case, M = 14 symbols × 12 subcarriers =168. The MSE and Crame´r-Rao bound for these two cases are shown in Figure 4.3aand Figure 4.3b respectively. In the third case, we have DMRS symbols spanning 12subcarriers transmitted every half-subframe, corresponding to L = 12, M = 12 and atotal of 2N DMRS transmissions. The results for this case are shown in Figure 4.3c.The SNR considered for this analysis is between -20 dB and -10 dB, correspondingto the operating scenarios for MTC UEs in low network coverage. It can be observedthat with increasing SNR, the MSE of our ML based CFO estimator gets closer tothe Crame´r-Rao bound for all the three cases. Also, the performance of the firsttwo cases is better than that of the third case due to larger value of M in thesecases. Moreover, the MSE is measured for actual CFO values, which means thatan estimation error of 10 Hz corresponds to MSE = 100. Therefore, for an AWGNchannel with SNR of -15.5 dB (corresponding to 18 dB coverage enhancement) andthe desired CFO estimation accuracy of 10 Hz, NCFO = 16 for the first two cases andNCFO = 32 for the third case.4.5.3 Results for the EPA ChannelNow, we analyze the performance of our ML based CFO estimation techniques forthe EPA channel. Owing to the very low operating SNR, achieving the desired accu-racy of estimation with 100% probability may take a very large number of repetitionsfor fading channels. In such cases, it is necessary to evaluate the performance basedon a probabilistic measure, i.e., achieving the desired accuracy of estimation withhigh probability. Therefore, we define the performance metric as the number of sub-frames (N) required to achieve CFO estimation error ≤ 10 Hz with 95% probability113Chapter 4. Low SNR Uplink CFO Estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF (DMRS compensated)32 SF (DMRS compensated)64 SF (DMRS compensated)128 SF (DMRS compensated)16 SF32 SF64 SF128 SF(a) Using ML estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF (DMRS compensated)32 SF (DMRS compensated)64 SF (DMRS compensated)128 SF (DMRS compensated)16 SF32 SF64 SF128 SF(b) Using angle-based estimationFigure 4.4: CDF of the estimated CFO error using RV repetitions for legacyLTE/LTE-A uplink.114Chapter 4. Low SNR Uplink CFO Estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF (DMRS compensated)32 SF (DMRS compensated)64 SF (DMRS compensated)128 SF (DMRS compensated)16 SF32 SF64 SF128 SF(a) Using ML estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF (DMRS compensated)32 SF (DMRS compensated)64 SF (DMRS compensated)128 SF (DMRS compensated)16 SF32 SF64 SF128 SF(b) Using angle-based estimationFigure 4.5: CDF of the estimated CFO error using RV repetitions for MTCLTE/LTE-A uplink.115Chapter 4. Low SNR Uplink CFO Estimationand present the results in terms of the Cumulative Distribution Function (CDF) ofthe CFO estimation error for different values of N . A similar approach is adoptedby different companies in the 3GPP when they provide the performance results fordownlink synchronization, where the number of repetitions of the synchronizationsignal required to achieve 90% detection probability is used as the performance met-ric [43, 44].For the first two cases, we compare the results obtained by our method with theCFO estimation scheme using the conventional angle-based scheme (see Eq. (4.4)).For the DMRS only case, we compare the results from the ML based estimationscheme with that obtained from the modified angle-based scheme (see Eq. (4.17)).The simulation settings are summarized in Table 4.1.Figure 4.4a and Figure 4.4b indicate the performance of our ML based CFO es-timation algorithm and the conventional angle-based CFO estimation algorithm re-spectively, for the legacy RV transmission scheme (RV pattern: 0, 2, 3, 1, 0, 2, 3, 1, · · · )in LTE/LTE-A uplink. We observe that our ML estimation based method requiresat least NCFO = 64 for ≥ 95% probability of successful CFO estimation, while forNCFO = 16, this probability reduces to 68%. Therefore, for the desired CFO perfor-mance (95% success rate with 10 Hz accuracy), the eNB has to buffer 64 subframes.The conventional angle-based method has only 80% success rate in CFO estimationeven when NCFO = 128. In both the cases, using the entire subframe with DMRScompensation performs marginally better than using only the data symbols becausewe have 14 symbols instead of 12 available for CFO estimation.Figure 4.5a and Figure 4.5b depict the performance the two CFO estimationalgorithms for the MTC RV transmission scheme (RV pattern with N RV-0s, followedby N RV-2s and so on) in LTE/LTE-A uplink. Since the CFO is estimated using N116Chapter 4. Low SNR Uplink CFO Estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF32 SF64 SF128 SF(a) Using ML estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF32 SF64 SF128 SF(b) Using modified angle-based estimationFigure 4.6: CDF of the estimated CFO error using DMRS only for both legacy andMTC LTE/LTE-A uplink.117Chapter 4. Low SNR Uplink CFO Estimationsubframes, NCFO = N . In this case, we observe that our ML estimation based methodwith NCFO = 16 has around 82% probability of successful CFO estimation, which isa significant improvement when compared to the legacy case, while for NCFO = 32,we see that the probability of successful estimation increases to 95%. This meansthat the eNB has to buffer 32 subframes for CFO estimation and correction, whichis half the size required for the legacy RV transmission scheme. The conventionalangle-based method has similar performance to that of the legacy case. Therefore,for both the legacy and MTC RV repetition schemes, the correlation phase anglemethod fails to achieve the same estimation accuracy as that of the ML based CFOestimator.Figure 4.6a and Figure 4.6b show the CDF of the CFO estimation error usingour ML estimation based method and the conventional angle-based method usingonly the DMRS signals. Also, there is no need to differentiate the legacy and MTCscenarios since the DMRS transmission mechanism remains the same for both thescenarios. We observe that the conventional method fails to provide an accurate CFOestimation even with averaging over 128 subframes because there are only 2 symbolsavailable per subframe for CFO estimation. Also, our ML based CFO estimationtechnique using 2 DMRS symbols performs as well as the same technique for legacyRV scheme, requiring N = 64 subframes for estimating the CFO within 10 Hz with95% probability. Using 2 DMRS symbols per subframe is as good as using 12 datasymbols because the noise on the DMRS symbols is averaged 2N times, while thaton data symbols is averaged N times, resulting in better performance. Although thismeans that we use only one-seventh of the symbols for CFO estimation (2 DMRSsymbols instead of 14 symbols in the legacy RV scheme), the eNB still needs to bufferthe entire 64 subframes, since the CFO correction has to be applied on all the data118Chapter 4. Low SNR Uplink CFO Estimationsymbols.To this end, we have shown that the reduction in residual CFO results in anincrease in the energy efficiency of the MTC UE (see Table 4.2) and that our proposedML based CFO estimation techniques provide a robust and an accurate mechanismto reduce the residual CFO in low coverage. Also, the number of subframes requiredby our technique for CFO detection is smaller than the total number of subframesrequired for successful data decoding (see Table 4.2), i.e., NCFO < NSF, ensuringthe feasibility of implementation of our technique at the eNB. In the following, wego one step further and apply our improved methods to the so-called large transportblock transmission mechanism for MTC UEs in LTE/LTE-A, to further enhance theirenergy efficiency.4.5.4 NB-IoT Large Transport Block TransmissionIn the current LTE/LTE-A standards, the maximum TBS is fixed based on theMCS and the number of PRBs allocated to the UE. With QPSK chosen to be thehighest order of modulation for NB-IoT, the maximum TBS that can be transmittedcorresponds to MCS = 9, which is 136 bits [57, 58]. The 3GPP standardizationactivities are considering a large transport block transmission mechanism, where theUE transits transport blocks whose size is larger than the current maximum size of136 bits. Now, we briefly review the large transport block transmission mechanismand demonstrate the energy efficiency gains when our CFO estimation technique isapplied to such a mechanism.The large transport block transmission mechanism relies on the precedent thatthe effective data rate of the UE increases when larger sized blocks are transmitted,which means that the UE can complete its transmission quickly, go back to idle state119Chapter 4. Low SNR Uplink CFO Estimationand save power. The effective data rate of the UE is given by Eq. (1.1). If we increasethe TBS by a factor α in Eq. (1.1), then NSF need not increase by the same amount.This is because the performance of the turbo decoder used for decoding the transportblock does not vary linearly with respect to the code rate. In most of the cases, NSFwill scale by a factor less than α, thereby increasing the effective data rate. However,the TBS cannot be increased arbitrarily and is limited by the code rate.The transport block is appended with a 24-bit CRC [58, 59]. The code rate persubframe is calculated ascorig =(TBS + 24)(Tsc × nb) (4.18)where Tsc denotes the total number subcarriers available for transmitting the trans-port block and nb is the number of bits per subcarrier. Since, MTC transmission isrestricted to QPSK, nb = 2. Considering that the uplink transmission for a singleantenna UE requires 2 symbols for DMRS transmission (see Figure 1.4), there are 12symbols available for control and data transmission. Assuming that the UE uses asingle PRB pair transmission (12 subcarriers per symbol) and does not transmit anycontrol information when it is sending data, Tsc = 12× 12 = 144.Legacy LTE/LTE-A standards indicate that the transport block transmissionshould obey the condition of each RV being independently decodable [58]. Hence,the code rate must be chosen per RV. However, this condition is relaxed for low-complexity MTC UEs, since they are low data-rate devices and require multipleretransmissions of data in most of the cases. Therefore, we have a new metric calledthe effective code rate, which is a measure of the code rate over 4 RVs, given byceff =corig4and the data block can be decoded when all 4 RVs are received.To illustrate this aspect, let us choose TBS = 324 bits. Noting that Tsc = 144 andnb = 2, we get corig = 1.21(> 1), which suggests that each RV transmission will not120Chapter 4. Low SNR Uplink CFO EstimationTable 4.3: Number of repetitions vs. CFOTBS N100Hz N10Hz R100Hz R10Hz(kbps) in (kbps)72 110 80 0.44 0.60144 200 144 0.60 0.83224 304 216 0.66 0.93328 376 256 0.81 1.19424 448 304 0.89 1.32be independently decodable for this TBS and it cannot be used in the legacy scheme.However, for the MTC scheme, the effective code rate ceff = 0.3(< 1), which suggeststhat the TBS is can be readily used.4.5.5 Energy Efficiency AnalysisNow, we illustrate the reduction in energy consumption of MTC UEs obtained bythe use of large transport block transmission and our ML based CFO estimation. Weuse the energy consumption model described in Section 4.2.1 to calculate the energyefficiency of the MTC UEs. The simulation parameters are listed in Table 4.1 andthe power consumption values used are the same as in Section 4.2.2.Table 4.3 gives the number of subframes required by the eNB to decode thetransport block (N100Hz and N10Hz) and the effective data rate for large transportblock transmission (R100Hz and R10Hz) with CFO = 100 Hz and CFO = 10 Hz,respectively. The former CFO value is the one currently used by the 3GPP for MTCperformance evaluation and the latter models the negligible frequency offset scenario(refer Section 4.2.2). Similar to the performance of the regular sized transport block(TBS = 72 bits) in Section 4.2.2, we observe that the effective data of the UE increaseswith a decrease in the residual CFO value even for large transport blocks.121Chapter 4. Low SNR Uplink CFO EstimationTable 4.4: Energy efficiency gain vs. TBSTBS η1 η2 with p = 0.95D = D = D = D = D = D =10% 1% 0.1% 10% 1% 0.1%72 - - - 25.8% 25.6% 22.5%144 27.2% 26.9% 23.7% 45.2% 44.6% 39.3%224 33.6% 33.2% 29.3% 50.2% 49.5% 43.7%328 46.0% 45.3% 40.0% 60.0% 59.2% 52.3%424 51.1% 50.4% 44.5% 63.4% 62.5% 55.2%Table 4.5: Battery lifetime gain vs. TBS for transmitting 10 kilobytesTBS β1 β2Tx:Rx Power consumption share70:30 50:50 30:70 70:30 50:50 30:7072 - - - 1.19 - 1.22 1.13 - 1.15 1.07 - 1.08144 1.2 - 1.24 1.13 - 1.16 1.08 - 1.09 1.38 - 1.46 1.24 - 1.29 1.13 - 1.16224 1.26 - 1.31 1.17 - 1.2 1.1 - 1.11 1.44 - 1.54 1.28 - 1.33 1.15 - 1.18328 1.39 - 1.47 1.25 - 1.30 1.14 - 1.16 1.58 - 1.72 1.35 - 1.43 1.19 - 1.22424 1.45 - 1.56 1.29 - 1.34 1.15 - 1.18 1.63 - 1.8 1.38 - 1.46 1.2 - Large Transport Block Transmission OnlyFirst, we calculate the energy efficiency obtained solely by the use of large trans-port block transmission, where the residual CFO is not compensated and remainsat 100 Hz. Let Eltb denote the energy consumed in this scenario. Eltb is obtainedby using tON = tltb =1R100Hzs per bit in Eq. (4.2). The energy efficiency gain iscalculated asη1 =Eorig − EltbEorig= 1− tltb + v(tTotal − tltb)torig + v(tTotal − torig) (4.19) Large Transport Block Transmission with CFO EstimationWhen our ML based CFO estimation techniques are used, the corresponding energyconsumption, Ecfo is obtained by using tON = tcfo =1R10Hzs per bit in Eq. (4.2).However, the CFO estimation is successful with probability p owing to the low oper-ating SNR and limited symbol buffer size available at the eNB. Therefore, the energy122Chapter 4. Low SNR Uplink CFO Estimationconsumption of the UE with CFO estimation is given byEfinal = pEcfo + (1− p)Eorig (4.20)Then, the energy efficiency gain of the UE using CFO estimation is calculated asη2 =Eorig − EfinalEorig= p(1− EcfoEorig)= p(1− tcfo + v(tTotal − tcfo)torig + v(tTotal − torig))(4.21)We use the energy consumed by the UE for TBS = 72 bits and CFO = 100 Hz,Eorig, as our reference and evaluate the energy efficiency of different sized transportblocks transport blocks with CFO = 100 Hz and CFO = 10 Hz. Table 4.4 summarizesthe energy efficiency results for three different duty-cycles (D), corresponding to 10%,1% and 0.1% (refer Section 4.2.2).We observe that solely the large transport block (without CFO estimation) resultsin 23.7% to 44.5% more energy efficiency than the current mode of operation withTBS = 72 bits even for a very low duty cycle of 0.1%. When our ML based CFOestimation techniques are used, the residual CFO is within 10 Hz with 95% probability(p = 0.95). With this, we obtain a further improved energy efficiency of 39.3% to55.2% for larger TBS, indicating that robust CFO estimation at the eNB significantlyreduces the energy consumption of the MTC UEs in low coverage.Alternatively, the performance of our mechanisms can be analyzed using the bat-tery lifetime gain metric (similar to that discussed in Chapter 3, Section 3.4.3). Sinceour solutions here are with respect to UE transmission, the battery lifetime gain isgiven byβi =1(1− γi)sTx + sRx (4.22)123Chapter 4. Low SNR Uplink CFO Estimationwhere i = 1 for solely the large transport block transmission and i = 2 for the largetransport block transmission with CFO estimation. The parameter γi represents theenergy efficiency gain calculated for the transmission of a data packet of size ds bits.That is, γ1 is obtained by calculating η1 using tltb = ceil(dsR100Hz)and γ2 is obtainedby calculating η2 using tcfo = ceil(dsR10Hz). In both the cases, torig = ceil(dsRˆ100Hz),where Rˆ100Hz corresponds to R100Hz value of the TBS used as reference.Since γi denotes the energy efficiency gain of the UE transmitter for the transmis-sion of ds bits, the transmission energy consumed is (1 − γi)sTx. Table 4.5 providesthe battery lifetime gains for different TBS values for ds = 10 kilobytes using TBS =72 bits at CFO = 100 Hz as reference. For each entry, the lower value correspondsto D = 1% and the higher value corresponds to D = 10%.The results are similar to those obtained for energy efficiency gain in Table 4.4.Smaller duty cycles have lower gains and CFO estimation improves the battery life-time gain for the large transport block transmission. For example, with the largestTBS of 424 bits and Tx:Rx share of 50:50, we obtain a battery lifetime gain of 1.29 to1.34, which means that a battery lasting for 10 years would now last for 12.9 to 13.4years. With our robust CFO estimation and compensation, the battery would lastfor an improved lifetime of 13.8 to 14.6 years. Also, we obtain increased gains whenthe UE transmission has a higher power consumption share than reception, which isopposite to the results in Chapter 3 (see Table 3.3). However, this is expected becauseour mechanisms discussed here only reduce the transmitter energy consumption anddo not modify the receiver energy consumption.124Chapter 4. Low SNR Uplink CFO Estimationx = |Actual CFO - Estimated CFO| in Hz0 5 10 15 20 25 30F(x) SF32 SF64 SF128 SFFigure 4.7: CDF of the estimated CFO error using ML estimation and 2x DMRS.4.6 DiscussionThe ML based CFO estimation scheme using RV repetition for MTC UEs (seeFig 4.5a), which demonstrated the best performance, suggests that the eNB requires32 consecutive repetitions of the same RV for the desired CFO estimation perfor-mance. It would be beneficial to have an MTC transmission scheme, which not onlyassists in CFO estimation, but also removes the constraints on RV block transmis-sion and repetition. In this section, we propose a new uplink transmission schemewith increased DMRS density, which achieves this objective. Also, we briefly discusshow our ML estimation based CFO estimation mechanisms can be used in non-LTEscenarios.125Chapter 4. Low SNR Uplink CFO Estimation4.6.1 Increased DMRS Density Scheme in LTE/LTE-AUplinkIn the following, we propose a new transmission scheme for LTE/LTE-A uplink, wherethe DMRS density is doubled for N initial subframes and evaluate the performance ofour ML based CFO estimation technique using the DMRS technique. In the currentLTE/LTE-A uplink, the DMRS sequences are transmitted on the fourth and theeleventh symbols of a subframe with normal CP (see Figure 1.4). For our proposedtransmitted scheme, the MTC UEs double the DMRS density by transmitting newDMRS sequences on the third and the tenth symbols along with the legacy DMRSsequences for the initial N subframes and then reverts back to the legacy scheme.Figure 4.7 gives the performance our ML based CFO estimation scheme when theDMRS density is doubled. We observe that we can estimate the CFO within 10 Hz ofthe actual value with 95% probability when the accumulation time is 32 ms or more.The performance results are close to that of the ML based CFO estimation schemeusing RV repetition for MTC UEs (see Fig 4.5a) and the doubled DMRS densityscheme does not impose any restriction on the RV block being transmitted and thenumber of repetitions, as desired.The only disadvantage of this scheme is that there is an overhead of 2 symbolsfor first N subframes for each transmission. For example, with N = 32, we havean overhead of 64 symbols. With 14 symbols per subframe (1 subframe = 1 ms),the overhead time is less than 5 ms. Since, the transmission takes more than 100subframes for any TBS, the overhead is less than 5% for all the cases. Moreover, theeNB can utilize the increased DMRS density for better channel estimation, whichimproves the overall performance of data decoding and further reduces the overhead.Alternatively, one could use the advantage of better channel estimation to transmit126Chapter 4. Low SNR Uplink CFO Estimationdata with a higher MCS, thereby increasing the throughput. Therefore, for the samenumber of buffered subframes, our increased DMRS density based estimation is morebeneficial than estimation using repeated data.4.6.2 Application of ML Based CFO Estimation toNon-LTE ScenariosHitherto, we designed and developed CFO estimation mechanisms specific to theLTE/LTE-A frame structure considering RV repetitions and DMRS transmissions.However, this technique can be readily extended to any communication mechanismincorporating periodic data and/or pilot repetitions. The ML based CFO estimationfor such scenarios can be derived by choosing the appropriate values of K and L basedon the length and the periodicity of the repeated data/pilot signals in Eq. (4.14)(similar to how we derived the DMRS based estimation as a special case).4.7 ConclusionIn this chapter, we addressed the problem of improving the uplink energy efficiencyof MTC devices adhering to the NB-IoT framework. We showed that the energyefficiency of the MTC UE increases if the eNB adopts CFO estimation mechanismsthat reduce the residual CFO to negligible limits. We proposed an ML based CFOestimation mechanism that uses the data and pilot repetitions in LTE/LTE-A andillustrated that it significantly outperforms the legacy CFO estimation techniqueusing the phase of the correlation between consecutive data repetitions. We demon-strated that incorporating our ML based CFO estimation technique at the eNB results22.5%-55.2% reduction in energy consumption of the MTC UEs, when compared to127Chapter 4. Low SNR Uplink CFO Estimationthe case where the residual CFO is not compensated. We also proposed a variationof the LTE/LTE-A frame structure incorporating additional pilot signals during theinitial MTC transmissions, which assists in faster CFO estimation at the eNB withminimal overhead.Our ML based CFO estimation technique provides a robust mechanism to estimatethe CFO in low coverage. Moreover, it is applicable to all categories of LTE MTC UEsand can also be easily adopted to other wireless communication standards. Recently,NB-IoT standardization decided not to adopt the increased DMRS density scheme,in order to maintain full compatibility with legacy UEs. However, the prospects ofhaving such a scheme for 5G NR standardization is open for study.128Chapter 5Conclusions and Directions forFuture WorkThe research work presented in this thesis focused on developing mechanisms to im-prove the current LTE/LTE-A standards to effectively support the IoT. Specifically,we suggested enhancements to the MTC framework in LTE/LTE-A addressing theenergy efficiency of the UEs in normal coverage, as well as those requiring coverageenhancement. The solutions developed for normal coverage are applicable to the IoTscenarios of pet tracking or weather sensing in which the UEs have low-mobility andare located in regions where the network coverage is good. The mechanisms devel-oped for extended coverage are suitable for the IoT applications of patient healthmonitoring or smart metering, where the IoT devices are located in interiors of build-ings where the network coverage is low. Most importantly, our solutions are in linewith the standardization activities for MTC in 3GPP LTE/LTE-A to facilitate IoTand have minimal influence on the legacy UEs.In Chapter 2, we discussed the limitations of the DRX procedure, which is cur-rently used by the LTE/LTE-A UEs in the downlink for saving power. We showedthat the power savings obtained from the DRX mechanism depend on the durationfor which the UE is awake, i.e., the ON time of the UE, which in turn depends onthe procedure followed by the UE for decoding the paging information. We discussedthat the paging operation was computationally intensive and was performed every129Chapter 5. Conclusions and Directions for Future Worktime the UE wakes up from the DRX cycle, regardless of whether it is paged ornot. To simplify the paging decode procedure, we introduced the a modified DRXmechanism incorporating quick sleeping. Our Quick Sleeping Indication (QSI) wouldindicate whether the UE is receiving a valid page or not in the upcoming pagingtransmission. The UE would first decode the QSI and if it indicated an impendingpaging message, the UE would stay awake for decoding the page. However, if thereis no valid page, the UE would immediately go back to sleep and save energy.For the normal coverage case, we designed the quick sleeping mechanisms usingthe resources that were already being allocated by the base station, thereby avoidingresource allocation overhead. Specifically, we choose to transmit the QSI on thosephysical channels whose locations on the subframe grid are fixed - the broadcastor the synchronization channels. Then, we identified that the timing reacquisitionprocess further increased the ON time, especially when the coverage is low, leadingto increased energy consumption. To alleviate this problem, we introduced a robustQSI signal using the data channel. Although this signal required additional resources,it helped the UEs in low coverage to obtain both the paging and timing informationin parallel, which reduced the ON time and paging decoding complexity, therebysaving energy. Our DRX with quick sleeping solutions demonstrated more than 45%improvement in energy efficiency.In the aforementioned DRX with quick sleeping mechanism, the timing resolu-tion considered was at the symbol level (the timing was designated to be correctif the UE finds the correct symbol number). In Chapter 3, we first showed thatthe paging decode operation is very sensitive to timing offset and any deviation intiming beyond the cyclic prefix length degrades the decoding process. In order toobtain accurate sample-level timing detection, we explored the conventional timing130Chapter 5. Conclusions and Directions for Future Workacquisition algorithms - cyclic prefix autocorrelation and the synchronization signaldetection. We demonstrated that the conventional mechanisms require substantialON time for reacquiring the symbol timing within tolerable limits and hence lead toincreased energy consumption.To mitigate this problem, we introduced the enhanced Primary SynchronizationSignal (ePSS) within the LTE/LTE-A standardization framework and proposed anovel DRX mechanism which uses our ePSS for faster timing resynchronization andreduced energy consumption. We described two simple methods to design the ePSSsignal - by using multiple legacy PSS sequences and by using longer ZC sequences andindicated that our ePSS design requires less than 1.5% network resource overhead forbase-station bandwidths of 5 MHz or more. We also illustrated that the modifiedDRX mechanism using ePSS as QSI would result in 1.2 to 1.8 times improvement inthe battery life of the low-coverage, low-complexity MTC devices.Having described energy efficient mechanisms for MTC UEs in the downlink inChapter 2 and Chapter 3, the problem of improving the energy efficiency of the low-coverage MTC devices in the uplink was addressed in Chapter 4. We showed thatthe energy efficiency of the MTC UE increases if the base station adopts CFO esti-mation mechanisms that reduce the residual CFO to negligible limits. We proposeda ML based CFO estimation mechanism that uses the data and pilot repetitions inLTE/LTE-A and illustrated that it significantly outperforms the legacy CFO estima-tion technique using the phase of the correlation between consecutive data repetitions.An important contribution from this work was the evaluation of our CFO esti-mation algorithm for the large transport block transmission in the NB-IoT. Variousstudies in NB-IoT indicated that the large transport block transmission increasesthe effective data rate and reduces the retransmission time, thereby saving energy.131Chapter 5. Conclusions and Directions for Future WorkWe showed that incorporating our ML based CFO estimation mechanism results infurther reduction in the number of retransmissions for NB-IoT. Finally, we proposeda variation of the LTE/LTE-A frame structure incorporating additional pilot signalsduring the initial MTC transmissions, which assists in faster CFO estimation at theeNB with minimal overhead.From the standardization perspective, our DRX with QSI and ePSS solutions areapplicable to UE categories of CAT-M1, CAT-0 and above, since they operate on abandwidth of at least 1.4 MHz. These contributions were presented in the MTC RadioAccess network - Layer 1 (RAN-1) meetings of the 3GPP and played a vital role in - a)extending the DRX cycle length beyond the initially proposed 2.56 s to 2621.44 s forCAT-M1 UEs, b) developing a simplified paging control channel for CAT-M1 deviceswith a single Downlink Control Information (DCI) format to reduce the number ofblind hypotheses and c) identifying that UE resynchronization also plays a majorrole in increased energy consumption of low coverage MTC UEs. In fact, the topic ofimproving resynchronization (and initial synchronization) is still an open study itemfor LTE MTC Release 14. Recently, the focus of the LTE standardization has shiftedtowards defining NB-IoT and 5G NR specifications. Nevertheless, the learning andthe basic principles adopted in our solutions provide the impetus to design moreenergy efficient mechanisms in the downlink for the forthcoming NB-IoT and 5G NRstandards.Our ML based CFO estimation technique in the uplink is not only applicable toall the LTE UE categories, but also easily adoptable for other wireless communicationstandards using data/pilot repetitions. But our proposal for increased DMRS densitywas not considered for standardization, in order to maintain full compatibility withlegacy UEs. Considering that novel schemes for uplink pilot design are open study132Chapter 5. Conclusions and Directions for Future Workitems for 5G NR, we believe that it would be beneficial to develop these schemes suchthat they allow for faster and robust CFO estimation in low coverage.5.1 Directions for Future WorkThe IoT scenario and the MTC mechanisms present numerous challenging researchproblems. Some interesting avenues for future research are summarized below.5.1.1 Coverage Enhancement for LTE MTC Using MassiveMIMOMassive Multi-Input-Multi-Output (MIMO) is a communication technology that hasattracted the interest of researchers worldwide in recent years. It is a successorof the conventional MIMO technology which has been well studied and applied invarious wireless communication standards like WLAN (802.11), WiMax (802.16) andLTE. It has been proved that the MIMO technology improves the reliability andthe capacity of wireless communications significantly. The MIMO technology hasevolved from addressing point-to-point links where two devices with multiple antennascommunicate with each other, to providing efficient communication mechanisms forMulti-User MIMO (MU-MIMO) systems, where a base station with multiple antennascommunicates with a set of single antenna devices. The MU-MIMO system modelfits the case of MTC where the network comprises of multiple, low-complexity, singleantenna users. The current MU-MIMO systems employ few number of antennas (forexample, 8 in LTE) at the base station and hence extract relatively modest gainsin spectral efficiency. The massive MIMO technology employs a large number ofantennas at the base station (100 or more) in order to enhance the capacity and133Chapter 5. Conclusions and Directions for Future Workspectral efficiency of the system. In [120], it has been shown that massive MIMO canprovide a spectral efficiency as high as 26.5 bps/Hz.A prime advantage of massive MIMO is its energy efficient mode of operation whencompared to the conventional Single-Input-Single-Output (SISO) system. In [121], itis shown that the transmit power of each user in a massive MIMO system is reducedwhen compared to the SISO system for the same performance. This reduction inuser transmission power is proportional to M , the number of antennas at the basestation with perfect CSI or to√M in the case of imperfect CSI. This leads to energyefficient operation in the uplink. Similarly, in the downlink, the reduction in basestation transmit power is proportional to K and√K for perfect CSI and imperfectCSI respectively, where K is the number of users. In other words, massive MIMOcan provide extended coverage for the same amount of power. The MTC scenario isbound to have more than one user to be served at a given instant of time owing tothe large number of devices present in the network. This renders massive MIMO tobe a suitable candidate for coverage enhancement in MTC when some CSI (perfector imperfect) information is available.It would be interesting to examine the use of massive MIMO to improve theSNR during the initialization phase in LTE/LTE-A, where the CSI is completelyunknown. Recently, beamforming based initial access techniques are being proposedfor millimeter wave communications [122, 123]. While these works use the Line OfSight (LOS) propagation model for the channel and multiple antennas at the UE, itwould be important to evaluate the performance for single antenna UEs in Non LineOf Sight (NLOS) channels and also to standardize massive MIMO operation in 5G.Secondly, recent standardization activities for the 5G NR in LTE/LTE-A [124]discuss grant-free access methods for UE transmission [125]. These grant-free meth-134Chapter 5. Conclusions and Directions for Future Workods will be a perfect foil for massive MIMO base-stations, since the UEs can follow asimple transmission mechanism and the base station can incorporate improved detec-tion/decoding mechanisms of the UE signals using the large antenna array processing.For example, large antenna arrays can render better directivity for the base-stationto employ digital beamforming and process the UE signals one direction at a time,thereby mitigating the interference from the other directions. Developing massiveMIMO based grant-free mechanisms to address the massive uplink access in IoTwould also be a potential avenue for future research.5.1.2 Exploring LTE in Unlicensed Bands (LTE-U) forMTCWith the exponential increase in the number of devices requiring network access, itwould be highly difficult for the current frequency spectrum to provide resources forall these devices. Moreover, the purchase of a licensed spectrum is expensive and itsavailability is limited. Therefore, recent research has moved towards exploring unli-censed spectral bands, which has a large amount of available spectrum. For example,in the 5 GHz band, the available spectrum spans approximately 500 MHz [126]. Since,LTE/LTE-A provides better system capacity, higher coverage and easy integrationwith existing systems, deployment of LTE-Unlicensed (LTE-U) is being consideredby 3GPP. The LTE-A standard employs carrier aggregation to provide additionalbandwidth. In LTE-U, this idea is being extended towards Licensed Assisted Access(LAA) where the licensed and unlicensed bands are aggregated. Therefore, LTE-Ucan potentially combine the advantages of LTE-A with the large bandwidth availablein the unlicensed spectrum to handle the high data demand. This renders LTE-U tobe a good candidate for MTC.135Chapter 5. Conclusions and Directions for Future WorkThe main challenge for LTE-U is to co-exist with the other technologies likeWireless Fidelity (WiFi) and Bluetooth in the unlicensed band [127]. In this regard,regulatory mechanisms like Listen Before Talk (LBT), dynamic frequency selectionand Clear Channel Assessment (CCA) for the low-complexity MTC UEs have toinvestigated. A major advantage of using unlicensed band is data offloading and co-operative communications. LTE-U for cooperative communications is an appealingresearch topic, where the MTC UEs could use WiFi to exchange information amongstthemselves, while being connected to a primary LTE cellular network. One applica-tion of such a framework would be optimize the transmission of the Channel QualityIndication (CQI) in the uplink. The LTE-U framework can also be used along withthe massive MIMO solution, where cooperative CQI feedback mechanisms can beemployed to improve the system performance.5.2 Concluding RemarksThe advent of IoT and the emergence of a plethora of new challenges has not onlymade the field of wireless communication exciting, but also rekindled the interest indeveloping novel physical layer solutions adhering to the diverse IoT requirements.We believe that this is a great time to contribute to this field with innovative andpractical solutions. The contributions from this thesis provide a stepping stone foranalyzing the performance of IoT devices adopting the LTE/LTE-A MTC mecha-nisms. The energy efficiency and extended coverage aspects evaluated in this workare identified to be key performance indicators for the design and development of nextgeneration wireless communication technologies hosting the IoT. Lastly, although thisresearch is closely tied to LTE/LTE-A, it is fairly straightforward to extend the ideasto other standards/technologies for facilitating the IoT.136Bibliography[1] A. Zanella, N. Bui, A. Castellani, L. Vangelista, and M. Zorzi, “Internet ofthings for smart cities,” IEEE Internet Things J., vol. 1, no. 1, pp. 22–32,2014.[2] L. Da Xu, W. He, and S. Li, “Internet of things in industries: A survey,” IEEETrans. Ind. Informat., vol. 10, no. 4, pp. 2233–2243, 2014.[3] N. Dlodlo and J. Kalezhi, “The Internet of things in agriculture for sustainablerural development,” in Proc. IEEE Int. Conf. on Emerging Trends in Networksand Computer Communications, 2015, pp. 13–18.[4] C. Kalalas, L. Thrybom, and J. Alonso-Zarate, “Cellular communications forsmart grid neighborhood area networks: A survey,” IEEE Access, vol. 4, pp.1469–1493, 2016.[5] L. Catarinucci, D. de Donno, L. Mainetti, L. Palano, L. Patrono, M. L. Ste-fanizzi, and L. Tarricone, “An IoT-aware architecture for smart healthcare sys-tems,” IEEE Internet Things J., vol. 2, no. 6, pp. 515–526, 2015.[6] G. Kortuem, F. Kawsar, V. Sundramoorthy, and D. Fitton, “Smart objects asbuilding blocks for the Internet of things,” IEEE Internet Comput., vol. 14,no. 1, pp. 44–51, 2010.[7] Ericsson, “Cellular networks for massive IoT (whitepaper),” Jan 2016.[8] “IEEE standard for local and metropolitan area networks–part 15.4: Low-ratewireless personal area networks (LR-WPANs),” IEEE Std 802.15.4-2011, 2011.[9] D. Raychaudhuri and N. B. Mandayam, “Frontiers of wireless and mobile com-munications,” Proc. of the IEEE, vol. 100, no. 4, pp. 824–840, 2012.[10] “IEEE standard for information technology–telecommunications and informa-tion exchange between systems–local and metropolitan area networks–specificrequirements–part 11: Wireless lan medium access control (MAC) and physicallayer (PHY) specifications,” IEEE Std 802.11-2012, 2012.[11] S. Aust, R. V. Prasad, and I. G. Niemegeers, “IEEE 802.11 ah: Advantages instandards and further challenges for sub 1 GHz Wi-Fi,” in IEEE Int. Conf. onCommun., 2012, pp. 6885–6889.137Bibliography[12] D. Astely, E. Dahlman, G. Fodor, S. Parkvall, and J. Sachs, “LTE release 12and beyond,” IEEE Commun. Mag., vol. 51, no. 7, 2013.[13] I. Hwang, B. Song, and S. S. Soliman, “A holistic view on hyper-dense het-erogeneous and small cell networks,” IEEE Commun. Mag., vol. 51, no. 6, pp.20–27, 2013.[14] A. Damnjanovic, J. Montojo, Y. Wei, T. Ji, T. Luo, M. Vajapeyam, T. Yoo,O. Song, and D. Malladi, “A survey on 3GPP heterogeneous networks,” IEEEWireless Commun., vol. 18, no. 3, pp. 10–21, 2011.[15] J. Ramiro and K. Hamied, Self-Organizing Networks (SON): Self-Planning,Self-Optimization and Self-Healing for GSM, UMTS and LTE. John Wiley &Sons, 2011.[16] M. Peng, D. Liang, Y. Wei, J. Li, and H.-H. Chen, “Self-configuration andself-optimization in LTE-Advanced heterogeneous networks,” IEEE Commun.Mag., vol. 51, no. 5, pp. 36–45, 2013.[17] Nokia, “LTE evolution for IoT connectivity (whitepaper),” 2016.[18] A. Rajandekar and B. Sikdar, “A survey of MAC layer issues and protocols formachine-to-machine communications,” IEEE Internet Things J., vol. 2, no. 2,pp. 175–186, 2015.[19] X. Liang, X. Li, M. Barua, L. Chen, R. Lu, X. Shen, and H. Luo, “Enabling per-vasive healthcare through continuous remote health monitoring,” IEEE Wire-less Commun., vol. 19, no. 6, pp. 10–18, 2012.[20] M. R. Palattella, M. Dohler, A. Grieco, G. Rizzo, J. Torsner, T. Engel, andL. Ladid, “Internet of things in the 5G era: Enablers, architecture, and businessmodels,” IEEE J. Sel. Areas Commun., vol. 34, no. 3, pp. 510–527, 2016.[21], “Low power wide area Internet of things (LPWA-IoT) marketforecasts to 2020 and MNO approaches,” Apr. 2016.[22] Sigfox,, Last retrieved on 15th May 2017.[23] Semtech,, Last retrieved on 15th May 2017.[24] J. Petajajarvi, K. Mikhaylov, A. Roivainen, T. Hanninen, and M. Pettissalo,“On the coverage of LPWANs: range evaluation and channel attenuation modelfor LoRa technology,” in Proc. IEEE 14th Int. Conf. on ITS Telecommun.,2015, pp. 55–59.[25] N. Sornin, M. Luis, T. Eirich, T. Kramp, and O. Hersent, “LoRaWAN specifi-cation V1,” LoRa Alliance, Tech. Rep., 2015.138Bibliography[26] Ingenu, “RPMA technology for the Internet of things,” Ingenu, Tech. Rep.,2016.[27] Telensa,, Last retrieved on 15th May 2017.[28] Weightless,, Last retrieved on 15th May 2017.[29] 802.11 Task Group, “Long range low power (LRLP) operation in 802.11: Usecases and functional requirements: Guidelines for par development,” IEEEP802.11, 2016.[30] ETSI, “Low throughput networks (LTN); use cases for low throughput net-works,” ETSI GS LTN 001 V1.1.1, Sep. 2014.[31] ——, “Low throughput networks (LTN); functional architecture,” ETSI GSLTN 002 V1.1.1, Sep. 2014.[32] ——, “Low throughput networks (LTN); protocols and interfaces,” ETSI GSLTN 003 V1.1.1, Sep. 2014.[33] R. Ratasuk, S. Iraji, K. Hugl, L. Wang, and A. Ghosh, “Performance of low-cost LTE devices for advanced metering infrastructure,” in Proc. IEEE 77thVeh. Tech. Conf.-Spring, 2013, pp. 1–5.[34] Y. Chen and W. Wang, “Machine-to-machine communication in LTE-A,” inProc. IEEE 72nd Veh. Technol. Conf.-Fall, 2010, pp. 1–4.[35] S.-Y. Lien, K.-C. Chen, and Y. Lin, “Toward ubiquitous massive accesses in3GPP machine-to-machine communications,” IEEE Commun. Mag., vol. 49,no. 4, pp. 66–74, 2011.[36] T. Taleb and A. Kunz, “Machine type communications in 3GPP networks:potential, challenges, and solutions,” IEEE Commun. Mag., vol. 50, no. 3, pp.178–184, 2012.[37] A. G. Gotsis, A. S. Lioumpas, and A. Alexiou, “M2M scheduling over LTE:Challenges and new perspectives,” IEEE Veh. Technol. Mag., vol. 7, no. 3, pp.34–39, 2012.[38] F. Ghavimi and H.-H. Chen, “M2M communications in 3GPP LTE/LTE-Anetworks: Architectures, service requirements, challenges, and applications,”IEEE Commun. Surveys Tuts., 2015.[39] J. J. Nielsen, G. C. Maduen˜o, N. K. Pratas, R. B. Sørensen, C. Stefanovic, andP. Popovski, “What can wireless cellular technologies do about the upcomingsmart metering traffic?” IEEE Commun. Mag., vol. 53, no. 9, pp. 41–47, 2015.139Bibliography[40] G. C. Maduen˜o, J. J. Nielsen, D. M. Kim, N. K. Pratas, Cˇ. Stefanovic´, andP. Popovski, “Assessment of LTE wireless access for monitoring of energy dis-tribution in the smart grid,” IEEE J. Sel. Areas Commun., vol. 34, no. 3, pp.675–688, 2016.[41] C.-H. Wei, R.-G. Cheng, and F. M. Al-Taee, “Dynamic radio resource allocationfor group paging supporting smart meter communications,” in Proc. IEEE 3rdInt. Conf. on Smart Grid Commun., 2012, pp. 659–663.[42] R. Ratasuk, A. Prasad, Z. Li, A. Ghosh, and M. A. Uusitalo, “Recent advance-ments in M2M communications in 4G networks and evolution towards 5G,” inProc. IEEE 18th Int. Conf. on Intelligence in Next Generation Netw., 2015,pp. 52–57.[43] 3GPP, “Study on provision of low-cost machine-type communications (MTC)user equipments (UEs) based on LTE,” 3GPP, TR 36.888, Jun. 2013.[44] ——, “Study on machine-type communications (MTC) and other mobile dataapplications communications enhancements,” 3GPP, TR 23.887, Dec. 2013.[45] R. Ratasuk, B. Vejlgaard, N. Mangalvedhe, and A. Ghosh, “NB-IoT systemfor M2M communication,” in Proc. IEEE Wireless Commun. and Netw. Conf.,2016, pp. 1–5.[46] 3GPP TSG RAN, “RP-152284, Revised Work Item: Narrowband IoT (NB-IoT),” 3GPP, Dec. 2015.[47] ——, “R1-156010, NB-LTE General L1 Concept Description,” 3GPP, Oct.2015.[48] 3GPP, “Cellular system support for ultra-low complexity and low throughputinternet of things (CIoT),” 3GPP, TR 45.820, Dec. 2015.[49] N. M. Balasubramanya, L. Lampe, G. Vos, and S. Bennett, “Introducing quicksleeping using the broadcast channel for 3GPP LTE MTC,” in Proc. IEEEGlobal Telecommun. Conf. Wkshps., 2014, pp. 606–611.[50] ——, “DRX with quick sleeping: A novel mechanism for energy-efficient IoTusing LTE/LTE-A,” IEEE Internet Things J., vol. 3, no. 3, pp. 398–407, 2016.[51] N. M. Balasubramanya, L. H.-J. Lampe, and G. G. Vos, “Quick paging methodand apparatus in LTE,” 2015, US Patent App. 14/799,513.[52] N. M. Balasubramanya, L. Lampe, G. Vos, and S. Bennett, “On timing reacqui-sition and enhanced primary synchronization signal (ePSS) design for energyefficient 3GPP LTE MTC,” IEEE Trans. Mobile Comput., 2016.140Bibliography[53] G. G. Vos, N. M. Balasubramanya, S. J. Bennett, and L. H.-J. Lampe, “Methodand system for providing and using enhanced primary synchronization signalfor LTE,” 2016, US Patent App. 15/010,192.[54] N. M. Balasubramanya, L. Lampe, G. Vos, and S. Bennett, “Low snr uplinkCFO estimation for energy efficient IoT using LTE,” IEEE Access, vol. 4, pp.3936–3950, 2016.[55] ——, “Method and system for carrier frequency offset estimation in LTE MTCdevice communication,” 2016, US Patent App. 15/010,192.[56] 3GPP, “Technical specification group services and system aspects; architectureenhancements to facilitate communications with packet data networks and ap-plications (release 13),” 3GPP, TS 23.682, Mar. 2016.[57] ——, “Evolved universal terrestrial radio access (E-UTRA); physical layer pro-cedures,” 3GPP, TS 36.213, Mar. 2014.[58] H. Holma and A. Toskala, LTE for UMTS: Evolution to LTE-Advanced. JohnWiley & Sons, 2011.[59] 3GPP, “Evolved universal terrestrial radio access (E-UTRA); medium accesscontrol (MAC) protocol specification,” 3GPP, TS 36.321, Mar. 2014.[60] ——, “Evolved universal terrestrial radio access (E-UTRA); user equipment(UE) procedures in idle mode,” 3GPP, TS 36.304, Mar. 2014.[61] ——, “Evolved universal terrestrial radio access (E-UTRA); physical channelsand modulation,” 3GPP, TS 36.211, Mar. 2014.[62] ——, “Evolved universal terrestrial radio access (E-UTRA); multiplexing andchannel coding,” 3GPP, TS 36.212, Dec. 2013.[63] J.-J. Van de Beek, M. Sandell, P. O. Borjesson et al., “ML estimation of timeand frequency offset in OFDM systems,” IEEE Trans. Signal Process., vol. 45,no. 7, pp. 1800–1805, 1997.[64] M. Speth, F. Classen, and H. Meyr, “Frame synchronization of OFDM systemsin frequency selective fading channels,” in Proc. IEEE 47th Veh. Technol. Conf.,vol. 3, 1997, pp. 1807–1811.[65] J.-B. Seo, S.-Q. Lee, N.-H. Park, H.-W. Lee, and C.-H. Cho, “Performanceanalysis of sleep mode operation in IEEE 802.16 e,” in Proc. IEEE 60th Veh.Technol. Conf.-Fall, vol. 2, 2004, pp. 1169–1173.[66] Y. Xiao, “Energy saving mechanism in the IEEE 802.16 e wireless MAN,” IEEECommun. Lett., vol. 9, no. 7, pp. 595–597, 2005.141Bibliography[67] Y. Zhang and M. Fujise, “Energy management in the IEEE 802.16 e MAC,”IEEE Commun. Lett., vol. 10, no. 4, pp. 311–313, 2006.[68] S. Baek, J. J. Son, and B. D. Choi, “Performance analysis of sleep mode op-eration for IEEE 802.16 m advanced WMAN,” in Proc. IEEE Int. Conf. onCommun. Wkshps., 2009, pp. 1–5.[69] K. De Turck, S. De Vuyst, D. Fiems, S. Wittevrongel, and H. Bruneel, “Per-formance of the sleep-mode mechanism of the new IEEE 802.16 m proposalfor correlated downlink traffic,” in Proc. Int. Conf. on Network Control andOptimization. Springer, 2009, pp. 152–165.[70] E. Hwang, K. J. Kim, J. J. Son, and B. D. Choi, “The power-saving mechanismwith periodic traffic indications in the IEEE 802.16 e/m,” IEEE Trans. Veh.Technol., vol. 59, no. 1, pp. 319–334, 2010.[71] S. Jin, M. Choi, and S. Choi, “Performance analysis of IEEE 802.16 m sleepmode for heterogeneous traffic,” IEEE Commun. Lett., vol. 14, no. 5, pp. 405–407, 2010.[72] S.-R. Yang and Y.-B. Lin, “Modeling UMTS discontinuous reception mecha-nism,” IEEE Trans. Wireless Commun., vol. 4, no. 1, pp. 312–319, 2005.[73] S.-R. Yang, “Dynamic power saving mechanism for 3G UMTS system,” MobileNetworks and Applications, vol. 12, no. 1, pp. 5–14, 2007.[74] S. Yang, M. Yoo, and Y. Shin, “Adaptive discontinuous reception mechanismfor power saving in UMTS,” IEEE Commun. Lett., vol. 11, no. 1, pp. 40–42,2007.[75] S.-R. Yang, S.-Y. Yan, and H.-N. Hung, “Modeling UMTS power saving withbursty packet data traffic,” IEEE Trans. Mobile Comput., vol. 6, no. 12, pp.1398–1409, 2007.[76] S.-r. Yang, P. Lin, and P.-t. Huang, “Modeling power saving for GAN andUMTS interworking,” IEEE Trans. Wireless Commun., vol. 7, no. 12, pp. 5326–5335, 2008.[77] C. S. Bontu and E. Illidge, “DRX mechanism for power saving in LTE,” IEEECommun. Mag., vol. 47, no. 6, pp. 48–55, 2009.[78] L. Zhou, H. Xu, H. Tian, Y. Gao, L. Du, and L. Chen, “Performance analysis ofpower saving mechanism with adjustable DRX cycles in 3GPP LTE,” in Proc.IEEE 68th Veh. Technol. Conf.-Fall, 2008, pp. 1–5.142Bibliography[79] J. Wigard, T. Kolding, L. Dalsgaard, and C. Coletti, “On the user performanceof LTE UE power savings schemes with discontinuous reception in LTE,” inProc. IEEE Int. Conf. on Commun. Wkshps., 2009, pp. 1–5.[80] S. Jin and D. Qiao, “Numerical analysis of the power saving in 3GPP LTEadvanced wireless networks,” IEEE Trans. Veh. Technol., vol. 61, no. 4, pp.1779–1785, 2012.[81] T. Kolding, J. Wigard, and L. Dalsgaard, “Balancing power saving and singleuser experience with discontinuous reception in LTE,” in Proc. IEEE Int. Symp.on Wireless Commun. Syst., 2008, pp. 713–717.[82] S. C. Jha, A. T. Koc, and R. Vannithamby, “Optimization of discontinuousreception (DRX) for mobile internet applications over LTE,” in Proc. IEEEVeh. Technol. Conf.-Fall, 2012, pp. 1–5.[83] Y. Y. Mihov, K. M. Kassev, and B. P. Tsankov, “Analysis and performanceevaluation of the DRX mechanism for power saving in LTE,” in Proc. IEEE26th Convention of Electrical and Electronics Engineers in Israel, 2010, pp.000 520–000 524.[84] S. Fowler, “Study on power saving based on radio frame in LTE wireless commu-nication system using drx,” in Proc. IEEE Global Telecommun. Conf. Wkshps.,2011, pp. 1062–1066.[85] S. Fowler, R. S. Bhamber, and A. Mellouk, “Analysis of adjustable and fixedDRX mechanism for power saving in LTE/LTE-Advanced,” in Proc. IEEE Int.Conf. on Commun., 2012, pp. 1964–1969.[86] A. T. Koc, S. C. Jha, R. Vannithamby, and M. Torlak, “Device power sav-ing and latency optimization in LTE-A networks through DRX configuration,”IEEE Trans. Wireless Commun., vol. 13, no. 5, pp. 2614–2625, 2014.[87] C.-C. Tseng, H.-C. Wang, F.-C. Kuo, K.-C. Ting, H.-H. Chen, and G.-Y. Chen,“Delay and power consumption in LTE/LTE-A DRX mechanism with mixedshort and long cycles,” IEEE Trans. Veh. Technol., vol. 65, no. 3, pp. 1721–1734, 2016.[88] K. Zhou, N. Nikaein, and T. Spyropoulos, “LTE/LTE-A discontinuous recep-tion modeling for machine type communications,” IEEE Wireless Commun.Lett., vol. 2, no. 1, pp. 102–105, 2013.[89] T. Tirronen, A. Larmo, J. Sachs, B. Lindoff, and N. Wiberg, “Reducing energyconsumption of LTE devices for machine-to-machine communication,” in Proc.IEEE Global Telecommun. Conf. Wkshps., 2012, pp. 1650–1656.143Bibliography[90] S. Kim, K. Jung, J. Choi, and Y. Kwak, “Improving LTE/LTE-A UE powerefficiency with extended DRX cycle,” in Proc. IEEE 80th Veh. Technol. Conf.-Fall, 2014, pp. 1–5.[91] S. Fowler, A. O. Shahidullah, M. Osman, J. M. Karlsson, and D. Yuan, “Analyt-ical evaluation of extended DRX with additional active cycles for light traffic,”Elsevier - Computer Networks, vol. 77, pp. 90–102, 2015.[92] S. C. Jha, A. T. Koc, and R. Vannithamby, “Device power saving mechanismsfor low cost MTC over LTE networks,” in Proc. IEEE Int. Conf. on Commun.Wkshps., 2014, pp. 412–417.[93] H. Chao, Y. Chen, and J. Wu, “Power saving for machine to machine commu-nications in cellular networks,” in Proc. IEEE Global Commun. Conf. Wkshps.,2011, pp. 389–393.[94] C.-Y. Tu, C.-Y. Ho, and C.-Y. Huang, “Energy-efficient algorithms and eval-uations for massive access management in cellular based machine to machinecommunications,” in Proc. IEEE Veh. Technol. Conf.-Fall, 2011, pp. 1–5.[95] J.-P. Cheng, C.-h. Lee, and T.-M. Lin, “Prioritized random access with dynamicaccess barring for RAN overload in 3GPP LTE-A networks,” in Proc. IEEEGlobal Commun. Conf. Wkshps., 2011, pp. 368–372.[96] S. Duan, V. Shah-Mansouri, and V. W. Wong, “Dynamic access class barringfor M2M communications in LTE networks,” in Proc. IEEE Global Commun.Conf., 2013, pp. 4747–4752.[97] M. Gerasimenko, V. Petrov, O. Galinina, S. Andreev, and Y. Koucheryavy,“Impact of machine-type communications on energy and delay performance ofrandom access channel in LTE-Advanced,” Wiley Online Library - Transactionson Emerging Telecommunication Technologies, vol. 24, no. 4, pp. 366–377, 2013.[98] M. Hasan, E. Hossain, and D. Niyato, “Random access for machine-to-machinecommunication in LTE-Advanced networks: issues and approaches,” IEEECommun. Mag., vol. 51, no. 6, pp. 86–93, 2013.[99] C. Y. Ho and C.-Y. Huang, “Energy-saving massive access control and resourceallocation schemes for M2M communications in OFDMA cellular networks,”IEEE Wireless Commun. Lett., vol. 1, no. 3, pp. 209–212, 2012.[100] S. Andreev, A. Larmo, M. Gerasimenko, V. Petrov, O. Galinina, T. Tirronen,J. Torsner, and Y. Koucheryavy, “Efficient small data access for machine-typecommunications in LTE,” in Proc. IEEE Int. Conf. on Commun., 2013, pp.3569–3574.144Bibliography[101] K. Wang, J. Alonso-Zarate, and M. Dohler, “Energy-efficiency of LTE for smalldata machine-to-machine communications,” in Proc. IEEE Int. Conf. on Com-mun., 2013, pp. 4120–4124.[102] H. S. Dhillon, H. C. Huang, H. Viswanathan, and R. A. Valenzuela, “Power-efficient system design for cellular-based machine-to-machine communications,”IEEE Trans. Wireless Commun., vol. 12, no. 11, pp. 5740–5753, 2013.[103] Y. Zhou, T. Zhang, Z. Zeng, Y. Li, and Y. Han, “LTE uplink coverage enhance-ment techniques based on enhanced TTI bundling,” in Proc. IEEE Int. Conf.on Wireless Commun., Netw. and Mobile Comput., 2012, pp. 1–4.[104] Y. Yuan, Z. Zuo, Y. Guan, X. Chen, W. Luo, Q. Bi, P. Chen, and X. She, “LTE-Advanced coverage enhancements,” IEEE Commun. Mag., vol. 52, no. 10, pp.153–159, 2014.[105] G. Naddafzadeh-Shirazi, L. Lampe, G. Vos, and S. Bennett, “Coverage en-hancement techniques for machine-to-machine communications over LTE,”IEEE Commun. Mag., vol. 53, no. 7, pp. 192–200, 2015.[106] 3GPP, “User equipment (UE) radio transmission and reception.” 3GPP, TS36.101, Mar. 2014.[107] W. Xu and K. Manolakis, “Robust synchronization for 3GPP LTE system,” inProc. IEEE Global Telecommun. Conf., 2010, pp. 1–5.[108] C.-L. Chen and S.-G. Chen, “Non-coherent cell identification detection methodsand statistical analysis for OFDM systems,” IEEE Trans. Commun., vol. 58,no. 11, pp. 3231–3243, 2010.[109] F. Berggren and B. M. Popovic´, “A non-hierarchical cell search scheme,” inProc. IEEE Wireless Commun. and Netw. Conf., 2007, pp. 2300–2304.[110] J.-W. Lee and Y.-H. Lee, “Rapid cell search in OFDM-based cellular systems,”in Proc. IEEE 61st Veh. Technol. Conf.-Spring, vol. 2, 2005, pp. 1273–1277.[111] P. Osti, P. Lassila, S. Aalto, A. Larmo, and T. Tirronen, “Analysis of PDCCHperformance for M2M traffic in LTE,” IEEE Trans. Veh. Technol., vol. 63,no. 9, pp. 4357–4371, 2014.[112] P. Frenger, H. Koorapaty, and J.-C. Guey, “Evaluation of control channel per-formance with adaptive radio unit activation in LTE,” in Proc. IEEE 75th Veh.Technol. Conf.-Spring, 2012, pp. 1–5.[113] J. Milos and S. Hanus, “Performance analysis of PCFICH and PDCCH LTEcontrol channels,” Radioengineering, vol. 23, no. 1, 2014.145Bibliography[114] J. Zhu and H. Li, “On the performance of LTE physical downlink shared chan-nel,” in Proc. IEEE Int. Conf. on Comput. Sci. and Netw. Technol., vol. 2,2011, pp. 983–986.[115] F. Riehle, Frequency standards: basics and applications. John Wiley & Sons,2006.[116] Qualcomm Incorporated, “R4-141241 - Reply LS on AGC and frequency errorfor D2D, GPP TSG-RAN WG2 Meeting 57bis, St. Julians, Malta 26 - 30 March2007.”[117] S. Ye, S. H. Wong, and C. Worrall, “Enhanced physical downlink control chan-nel in LTE Advanced Release 11,” IEEE Commun. Mag., vol. 51, no. 2, pp.82–89, 2013.[118] Nokia, “R2-071285 - DRX parameters in LTE, 3GPP TSG-RAN WG4 Meeting70, San Francisco, USA, 10th-14th Feb., 2014.”[119] M. Morelli and M. Moretti, “Robust frequency synchronization for OFDM-based cognitive radio systems,” IEEE Trans. Wireless Commun., vol. 7, no. 12,pp. 5346–5355, 2008.[120] L. Lu, G. Y. Li, A. L. Swindlehurst, A. Ashikhmin, and R. Zhang, “An overviewof massive MIMO: benefits and challenges,” IEEE J. Sel. Topics Signal Pro-cess., vol. 8, no. 5, pp. 742–758, 2014.[121] H. Q. Ngo, E. G. Larsson, and T. L. Marzetta, “Energy and spectral efficiencyof very large multiuser MIMO systems,” IEEE Trans. Commun., vol. 61, no. 4,pp. 1436–1449, 2013.[122] M. Giordani, M. Mezzavilla, and M. Zorzi, “Initial access in 5G mmwave cellularnetworks,” IEEE Commun. Mag., vol. 54, no. 11, pp. 40–47, 2016.[123] C. N. Barati, S. A. Hosseini, M. Mezzavilla, T. Korakis, S. S. Panwar, S. Ran-gan, and M. Zorzi, “Initial access in millimeter wave cellular systems,” IEEETrans. Wireless Commun., vol. 15, no. 12, pp. 7926–7940, 2016.[124] 3GPP, “Study on new radio (NR) access technology,” 3rd Generation Partner-ship Project (3GPP), TR 38.912, Sep. 2016.[125] C. Bockelmann, N. Pratas, H. Nikopour, K. Au, T. Svensson, C. Stefanovic,P. Popovski, and A. Dekorsy, “Massive machine-type communications in 5G:Physical and mac-layer solutions,” IEEE Commun. Mag., vol. 54, no. 9, pp.59–65, 2016.146Bibliography[126] R. Ratasuk, N. Mangalvedhe, and A. Ghosh, “LTE in unlicensed spectrumusing licensed-assisted access,” in Proc. IEEE Global Telecommun. Conf. Wk-shps., 2014, pp. 746–751.[127] F. M. Abinader, E. P. Almeida, F. S. Chaves, A. M. Cavalcante, R. D. Vieira,R. C. Paiva, A. M. Sobrinho, S. Choudhury, E. Tuomaala, K. Doppler et al.,“Enabling the coexistence of LTE and Wi-Fi in unlicensed bands,” IEEE Com-mun. Mag., vol. 52, no. 11, pp. 54–61, 2014.147Appendix AProof of Crame´r-Rao Bound inEq. (4.13)In this appendix, we derive the Crame´r-Rao bound for our ML based CFO estimatorfor repeated data. We begin with re-writing Eq. (4.9) for a given subcarrier m andthe Gaussian channel asY (m) = R(m)u(m) +W (m), (A.1)where Y (m) = [Y0(m), Y1(m), · · ·YN−1(m)]T is the received signal vector on the m-thsubcarrier and R(m) denotes the data transmitted on the m-th subcarrier. The CFOvector is given by u(m) = [1, e(j2piLKNs), e(j2pi2LKNs), · · · e( j2pi(N−1)LKNs )]T and the Gaussiannoise vector is indicated by W (m) = [W0(m),W1(m), · · ·WN−1(m)]T , which has zeromean and covariance matrix σ2(m)IN (IN is the identity matrix of order N). TheSNR is calculated asΨ =1MM−1∑m=0|R(m)|2σ2(m)(A.2)This is similar to the Eq. (4) in [119], for which the Crame´r-Rao bound is derivedusing the Fisher information matrix F (Eq. (42) in [119]). In our case, we obtainF =[D VV T βRHP−1R](A.3)148Appendix A. Proof of Crame´r-Rao Bound in Eq. (4.13)whereP = diag{σ2(0), σ2(1), · · · , σ2(M − 1)}, (A.4)D = N · diag{2P−1, 2P−1,P−2}, (A.5)V =2piLKN(N − 1)Ns· [−RTI P−1 RTRP−1 0TM ], (A.6)with RR and RI denoting the real and imaginary parts of R andβ =4pi2L2K2N(N − 1)(2N − 1)3N2s. (A.7)The Crame´r-Rao bound is given byCRB = [F−1]3M+1,3M+1 . (A.8)Similar to Eq. (47) in [119], for our case, we obtainb = α[RTI −RTR 0TM3NspiLK](A.9)as the last column of F−1, whereα =Ns2piLKN(N − 1)(4N − 3)RHP−1R . (A.10)Substituting Eq. (A.9) to Eq. (A.8) and simplifying, we getCRB =3N2s2pi2L2K2N(N − 1)(4N − 3)RHP−1R . (A.11)149Appendix A. Proof of Crame´r-Rao Bound in Eq. (4.13)Given thatRHP−1R =M−1∑m=0|R(m)|2σ2(m)= MΨ, (A.12)substituting Eq. (A.12) to Eq. (A.11), we obtain the final expression for the Crame´r-Rao bound given in Eq. (4.13).150


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items