Characterization of Persistent Multipath Components inIndoor and Outdoor Environment at 30 GHzbyBadrun Naher LiyaB. Sc. Engg., Khulna University of Engineering and Technology, 2011A THESIS SUBMITTED IN PARTIAL FULFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFMASTER OF APPLIED SCIENCEinTHE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES(Electrical and Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)September 2017c Badrun Naher Liya, 2017AbstractMillimeter-wave (mm-wave) frequency bands are under active consideration for use as short-rangemobile broadband links in fifth generation (5G) cellular access networks. Although channel char-acteristics such as path loss, delay spread and fading distributions have been extensively studied formm-wave channels, the study of the time-varying nature of the channel is still in its early stages.In this work, we studied the lifetime of multipath components of the mm-wave channel, usually re-ferred to as persistence. An important time-varying characteristic of the mm-wave channel, persis-tence may affect the capacity, and beam training and beam tracking process of mm-wave systems.We developed a 30-GHz vector-network-analyzer-based channel sounder suitable for characteriz-ing multipath persistence and verified its performance through a three-stage verification procedure;time and frequency domain verifications, two-ray verification, and measurements conducted usingthe National Institute of Standards and Technology (NIST) mm-wave channel sounder verificationartifact.The primary goal of this work was to characterize multipath persistence based on measurementsconducted at 30 GHz in indoor and outdoor urban microcell environments. Through analysis of ourmeasurement data, we confirmed that the log-logistic distribution provides an accurate descriptionof persistence and showed how the physical attributes of the channel influence the parameters ofthe distribution. We also verified that a weak correlation exists between average received powerand length of the persistent path. We further showed that the rate of angular change of a multipathcomponent throughout its lifetime follows a Laplace distribution and that the angular rate dependson the distance of reflectors from the transmitter-receiver path. We used these results to proposea simulation model that can be used to make simple ray tracing simulations more realistic and toassess the effect of persistence and variations in the angular rate on the capacity, and beam trainingand tracking process of mm-wave systems.iiLay SummaryIn the near future, the fifth generation (5G) of wireless technology will likely move beyond existingcellular bands below 6 GHz to include millimetre-wave bands above 30 GHz in order to achievemuch higher data rates and system capacities than ever before. Although such systems have greatpotential to improve subscriber access to high data rate services, their performance is strongly in-fluenced by the propagation impairments that signals encounter as they traverse the often complexpath from the transmitter to the receiver. The lifetime of multipath signals is often limited due tothe nature of reflecting or blocking objects. Here, we report our efforts to characterize multipathpersistence in indoor and outdoor environments over short distances including the lifetime andangular rate of individual multipath components and to develop a measurement-based simulationmodel that captures the essence of multipath persistence in a form useful in design and simulation.iiiPrefaceBadrun Naher Liya prepared this thesis under the supervision of Prof. David G. Michelson. Liyaand Dr. Michelson jointly designed the research plan and thesis organization. Liya conducted theliterature survey, wrote the Matlab scripts used for data analysis and generated the results. Prof.Michelson assisted with the writing and editing of the thesis.Dr. Siamak Bonyadi-Ram, Maryam Kashi, Anmol Bhardwaj, and Dr. Theo Mavridis con-tributed to the development of the mm-wave channel sounder and helped to collect the outdoormeasurement data. Candice Loo and Tran Mai Villageois assisted in the indoor measurement cam-paign and Visiting Prof. Mina Dashti contributed to making UBC building geodata usable throughproper modifications, which Liya used to generate ray tracing simulated data using the WirelessInSite propagation software.Portions of Chapter 3 were published in IEEE Globecom Workshops, 2016.B. N. Liya, and D. G. Michelson, “Characterization of multipath persistence in device-to-devicescenarios at 30 GHz,” in Proc. IEEE GlobecomWorkshops (GCWkshps), pp. 966–971, Dec. 2016.ivTable of ContentsAbstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiLay Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iiiPreface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ivTable of Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vList of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viiiList of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ixList of Abbreviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiiAcknowledgements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiv1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1 Significance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2 Previous Work and its Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51.4 Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52 Measurement System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.2 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92.3 Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122.4 Function and Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172.4.1 Instrument Connection and Configuration . . . . . . . . . . . . . . . . . . 172.4.2 Measurement Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.4.3 Visualize the measurement data in MATLAB . . . . . . . . . . . . . . . . 18v2.5 Calibration and Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182.6 Data Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232.6.1 Windowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252.6.2 Inverse Fast Fourier Transform (IFFT) . . . . . . . . . . . . . . . . . . . . 272.7 Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292.7.1 Verification in Time Domain of LOS Path . . . . . . . . . . . . . . . . . . 292.7.2 Verification in Frequency Domain of LOS Path . . . . . . . . . . . . . . . 322.7.3 Two Ray Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 342.8 Channel Sounder Verification by NIST . . . . . . . . . . . . . . . . . . . . . . . . 352.9 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Characterization of Multipath Persistence . . . . . . . . . . . . . . . . . . . . . . . . 393.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 393.2 Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.1 Physical Basis of Persistence . . . . . . . . . . . . . . . . . . . . . . . . . 423.2.2 Tracking and Modelling Approaches . . . . . . . . . . . . . . . . . . . . . 443.2.3 Persistent Path Tracking Algorithm . . . . . . . . . . . . . . . . . . . . . . 503.3 Measurement Campaign . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.3.1 Measurement System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.3.2 Measurement Scenarios . . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.3.3 Data Collection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553.3.4 Data Reduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.4.1 Distance Metric Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.4.2 Persistence Paths . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 673.4.3 Persistent Path Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . 683.4.4 Angular Rate Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 783.5 Application to Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 823.5.1 Simulation of Persistence and Random Scatterer . . . . . . . . . . . . . . . 823.5.2 Implementation of Realistic AOA Variation . . . . . . . . . . . . . . . . . 863.5.3 Implementation of Persistence . . . . . . . . . . . . . . . . . . . . . . . . 863.6 Impact of Persistence on System Performance . . . . . . . . . . . . . . . . . . . . 883.6.1 Throughput/Capacity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 883.6.2 Beam Training . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 893.6.3 Beam Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 903.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92vi4 Summary and Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 944.2 Limitations and Future Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96viiList of TablesTable 2.1 List of equipment. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14Table 2.2 Specifications of the experiment. . . . . . . . . . . . . . . . . . . . . . . . . . 14Table 2.3 Link budget. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15viiiList of FiguresFigure 2.1 (a) Photograph and (b) schematic diagram of our channel sounder. . . . . . . . 13Figure 2.2 Uncorrected measured channel frequency response of an LOS link (transmitter-receiver separation, 2 m). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16Figure 2.3 Power angular profile of received signal as displayed (transmitter-receiver sep-aration, 2m). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17Figure 2.4 Photograph of system response, Hb2b( f ) measurement. . . . . . . . . . . . . . 19Figure 2.5 (a) Frequency and (b) phase response of back-to-back system response, Hb2b( f ). 20Figure 2.6 (a) Frequency and (b) phase response of the RF cable. . . . . . . . . . . . . . . 21Figure 2.7 (a) Frequency and (b) phase response of the 30 dB attenuator-1. . . . . . . . . 22Figure 2.8 (a) Frequency and (b) phase response of an LOS link after removing the effectof the measurement system (transmitter-receiver separation, 2 m). . . . . . . . 24Figure 2.9 Channel impulse response of an LOS link with different window functions. . . 25Figure 2.10 Channel impulse response of an LOS link showing the effect of windowing. . . 26Figure 2.11 Channel impulse response (CIR) of an LOS link showing the effect of scaling. . 28Figure 2.12 Channel impulse response of an LOS link showing the effect of symmetric andnon-symmetric IFFT. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29Figure 2.13 (a) Channel impulse response of an LOS link (transmitter-receiver separation,2m) (b) Theoretical vs. actual path gain plot for LOS links (transmitter-receiverseparation, 2-16 m, with 2m steps). . . . . . . . . . . . . . . . . . . . . . . . . 30Figure 2.14 (a) Estimation of the transmitter-receiver separation from phase response and(b) Estimated vs. actual transmitter-receiver separation plot. . . . . . . . . . . 33Figure 2.15 Two-ray measurement scenario. . . . . . . . . . . . . . . . . . . . . . . . . . 34Figure 2.16 Channel impulse response of a NLOS reflected path. . . . . . . . . . . . . . . 35Figure 2.17 Setup of channel verification process using NIST channel sounder verificationartifacts. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36Figure 2.18 A typical verification result from the NIST channel sounder verification cam-paign. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37ixFigure 3.1 Physical basis of persistent paths. . . . . . . . . . . . . . . . . . . . . . . . . . 43Figure 3.2 AOA change for (a) large and (b) small separation between main reflector andtransmitter-receiver route. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45Figure 3.3 Time domain analysis of propagation paths (a) persistent paths with movingreceiver and (b) physical basis. . . . . . . . . . . . . . . . . . . . . . . . . . . 49Figure 3.4 Outdoor measurement sites. . . . . . . . . . . . . . . . . . . . . . . . . . . . 53Figure 3.5 Measurement sites, (a) without and (b) with vegetation. . . . . . . . . . . . . . 54Figure 3.6 Indoor measurement site (Hallway, wall to wall hallway width = 2.2m). . . . . 55Figure 3.7 (a) Received power distribution and (b) extracted MPCs for 3 m transmitter-receiver separation as a function of TOA and AOA. . . . . . . . . . . . . . . . 57Figure 3.8 Effect of transmitting antenna beam pattern. . . . . . . . . . . . . . . . . . . . 58Figure 3.9 Extracted MPCs of indoor scenario for receiver route 3-9 m (Hallway). . . . . . 60Figure 3.10 (a) Persistent paths considering 1D distance metric (AOA only) and (b) re-ceived power variation of a selected path (bold, red path) (Hallway). . . . . . . 61Figure 3.11 Two simple examples of received power variation of persistent paths from raytracing simulated data. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62Figure 3.12 (a) Persistent paths considering 2D distance metric (AOA and received power)and (b) received power variation of the selected path (bold, blue path) (Hallway). 64Figure 3.13 (a) Persistent paths considering 3D distance metric (AOA, TOA and receivedpower) and (b) received power variation of the selected path (Hallway). . . . . 66Figure 3.14 Micro persistent MPCs and dropouts. . . . . . . . . . . . . . . . . . . . . . . 67Figure 3.15 (a) PDF and (b) CDF of persistent paths (Hallway). . . . . . . . . . . . . . . . 69Figure 3.16 (a) Histogram of the shape factor, s of log-logistic distribution (Hallway) and(b) relationship between the complexity of the propagation channel and log-logistic shape factor, s , double circles are representing overlapped points. . . . 72Figure 3.17 The relation between shape factor of log-logistic distribution and distancethreshold (Hallway). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73Figure 3.18 Histograms of mean path length for, (a) 4m and (b) 2m step size. . . . . . . . . 75Figure 3.19 The relation between average received power and the length of the persistentpath at a specific scenario (Hallway). . . . . . . . . . . . . . . . . . . . . . . . 76Figure 3.20 (a) Histogram of the correlation coefficient of the average received power andthe length of the persistent path and (b) their relation with specular index, for12 different measurement sites, double circles are representing overlapped points. 77Figure 3.21 (a) PDF and (b ) CDF of Laplace distribution (Hallway). . . . . . . . . . . . . 79xFigure 3.22 (a) Laplace scale factor, b and the separation of the nearest wall from thetransmitter-receiver route and (b) Relationship between, Laplace scale factor,b and distance threshold, Thdist . . . . . . . . . . . . . . . . . . . . . . . . . . 81Figure 3.23 Persistent paths ray tracing simulation with (a) simplified parallel wall (b)a complex wall on one side and simplified wall on the other side of thetransmitter-receiver route. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83Figure 3.24 Persistent paths with (a) real AOA variation (b) real AOA variation and finitepersistence. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84Figure 3.25 Semi-Markov model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85Figure 3.26 Persistent paths for a real parallel wall scenario including persistent path andrandom scatterers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85xiList of Abbreviations1D one-dimensional2D two-dimensional3D three-dimensionalAHRS attitude heading reference systemAOA angle-of-arrivalAOD angle-of-departureAS angular separationBUC block-up-converterCDF cumulative distribution functionD2D device-to-deviceEKF extended Kalman filterIDFT inverse discrete Fourier transformIFFT inverse fast Fourier transformKS Kolmogorov-SmirnovLOS line-of-sightMCD multipath component distanceMIMO multiple-input/multiple-outputMPC multipath componentxiiNIST National Institute of Standard and TechnologyNLOS non line-of-sightOTA over-the-airOFDM orthogonal frequency-division multiplexingPDF probability distribution functionRMS root mean squareRF radio frequencySCPI Standard commands for programmable instrumentsSDMA space division multiple accessTOA time-of-arrivalUBC University of British ColumbiaUWB ultra wide bandVNA vector-network-analyzerWPAN wireless personal area networkZVD zero-velocity detectorxiiiAcknowledgementsFirst, I would like to thank my supervisor, Dr. David G. Michelson. Through his exceptionalmentorship, collaboration, and advice I have completed this study. I also appreciate his warm-hearted focus and support on his student’s best interests.I greatly appreciate the financial support that I have received during my course of study, includ-ing research assistantships sponsored by Huawei Technologies Canada and the Natural Science andEngineering Research Council of Canada, a Faculty of Applied Science Graduate Award from theUniversity of British Columbia, and a Vehicular Technology Travel Grant from the IEEE CanadianFoundation.I also would like to thank the members of Radio Science Lab, UBC and especially acknowl-edge the support of Dr. Siamak Bonyadi-Ram, Maryam Kashi, Anmol Bhardwaj, and Dr. TheoMavridis in developing the channel sounder and collecting outdoor measurement data. Thanks toCandice Loo and Tran Mai Villageois for their help in the indoor measurement campaign. Thanksto Visiting Prof. Mina Dashti for making UBC building geodata usable through proper modifica-tions, which we used to generate ray tracing simulated data. Special thanks to Mark Hawryluckand Dr. David Steer of Huawei Technologies Canada and Dr. Kate Remley and Dr. Jeanne Quimbyof NIST for offering valuable support, advice and suggestion during their visits to our lab.xivChapter 1Introduction1.1 SignificanceThe wireless industry is experiencing unprecedented challenges as it seeks to accommodate ex-plosive traffic growth within a limited wireless spectrum. To meet this growth, both academicsand industries have proposed to use mm-wave frequency as a fifth generation (5G) wireless accessscheme. Several groups, including the International Telecommunication Union (ITU) [1], Mo-bile and wireless communications Enablers for the Twenty-twenty Information Society (METIS)[2], Millimetre-Wave Based Mobile Radio Access Network for Fifth Generation Integrated Com-munications (mmMAGIC) [3], 5G mmWave Channel Model Alliance [4] are also working onmm-wave frequency bands. Of particular interest are the fixed/mobile bands at 28/30 and 38 GHz.The successful design, execution and performance assessment of mm-wave systems will requirethe understanding of the manner in which mm-wave signals are impaired as they traverse fromthe transmitter to the receiver because some signal propagation characteristics of mm-wave andexisting microwave frequencies are quite different.The interest for mm-wave reached to the peak in the recent years although it has been started awhile ago. Mm-wave frequency bands first have been researched at 57-64 GHz unlicensed band,to be used for wireless personal area network (WPAN) applications. To develop the required stan-dards for WPAN, IEEE 802.15.3 Task Group 3c (TG3c) was established in 2005. This task group1developed a millimetre-wave-based alternative physical layer (PHY) and defined beamforming andbeam searching protocols. They have also specified channel aggregation and block acknowledge-ment to improve the medium access control (MAC) efficiency at higher data rates [5]. However,because of the high development cost of 60 GHz equipment, we have not seen any successfulproduct based on 802.15.3c.Along with standardization organization, some commercial organization also came forward tomake the mm-wave technology feasible. The first such company is SiBEAM. SiBEAM has beenstarted by researchers at the University of California, Berkeley, to implement millimetre-wave(mm-wave) circuits on standard silicon chips. This company was the first to build 60GHz chipsetsusing standard CMOS technology, which has accelerated commercial interest in 60 GHz productsand helped to standardize the high-frequency mm-wave technology [6]. Lattice Semiconductorpurchased SiBEAM on June 1, 2015. The SiBEAM technology group at Lattice Semiconductorare carefully developing mm-wave RF components using CMOS technology.After successful implementation of 60 GHz chipsets by SiBEAM, wireless HD came to themarket. WirelessHD is the first industrial consortium, which defined a wireless video area network(WVAN) to transmit uncompressed HDMI signal over a 60 GHz frequency for the connection ofTV, DVD players, set-top boxes and portable devices. Although the high-frequency bands usuallyrequire line-of-sight (LOS) link between transmitter and receiver, the WirelessHD specificationaccomplished this by defining a protocol for directional connections that adapt very rapidly, usingbeam steering smart antennas and exploiting reflections and other non-LOS (NLOS) paths, whenthe LOS is obstructed [7]. They had developed an excellent MAC layer. Several products are cur-rently being advertised, including WirelessHD HDMI video transmitter-receiver and WirelessHDequipped televisions. However, due to technological limitations and cost the WirelessHD promotergroup has apparently shut down and Lattice semiconductor is implementing the technology.Another trade association, Wireless Gigabit Alliance (WiGig), came to the field to define spec-ifications and standards for high-frequency Wi-Fi. WiGig started work on two frequency bands,(i) below 6 GHz and (ii) above 6 GHz. Below 6 GHz came out as IEEE standard 802.11.ac and2above 6 GHz came out as IEEE standard 802.11.ad. IEEE standard 802.11.ad modifies both theIEEE 802.11 PHY layers and the IEEE 802.11 MAC layer to enable operation in frequenciesaround 60 GHz. Thus the technical contents of IEEE 802.11.ad standard is similar the Wi-Fistandard [8]. This standard also includes protocols for beam searching and steering technique. Al-though, some 802.11.ad products have already come to the market such as Intel tri-band antennamodule, but being very expensive still did not get much popularity.The above-mentioned mm-wave systems were intended for low-cost consumer products. Butmm-wave 5G cellular networks have a big support from industries, as many large companies suchas Apple, Samsung, Huawei, Ericsson, Nokia, are working on mm-wave to bring it to the reality asan alternative access scheme for 5G cellular networks. Along with working in their research anddevelopment group industries are also funding research organizations and academics to continuework on mm-wave channel model and system development to accelerate the process. The FederalCommunications Commission (FCC) has already approved 28 and 39 GHz bands for licensed cel-lular bands [9]. In addition to this, to release the first release of 5G specifications, standardizationorganization 3rd Generation Partnership Project (3GPP), recently planned a detailed work plan fortheir Release-15 [10]. All these efforts will accelerate the development of systems and productsfor mm-wave 5G cellular communication.1.2 Previous Work and its LimitationsBefore designing any system in a new frequency band, channel modelling is necessary to un-derstand the manner in which the propagation environment is impaired and distorted in dynamicscenarios. Channel models also play a significant role in testing and simulating wireless com-munication system. System designers require properly developed channel models to predict andcompare the performance of wireless communications systems under realistic conditions and toformulate and evaluate methods for mitigating the impairments and distortions that degrade wire-less signals.3High frequency, mm-wave propagation channel is very different from existing low-frequencypropagation channel, hence, PHY and MAC layers of a mm-wave system will be susceptible topropagation impairments in varying degrees. This difference between mm-wave and existing prop-agation channel necessitates proposing alternative PHY and MAC layers for the mm-wave systemand making both layers more robust. Channel models will provide a useful basis, against which al-ternative PHY or MAC layer proposals for use mm-wave systems can be evaluated and compared.Although channel modeling is a crucial stage for system development, most of the existingworks have been done for first-order channel parameters, such as path loss, delay spread, angularspread and K-factor [11–20]. However, understanding of second-order statistics is still inade-quate. We found a very few works, where second-order channel parameters, such as autocorrela-tion length [21] and Doppler spread [22] have been analyzed. Other channel fading parameters,such as Doppler rate, average fade duration (AFD), level crossing rate (LCR), and persistence areweakly represented in the literature. In cellular systems, one/both communication ends can bemobile. Even if both link ends are static, the propagation environment may change with time.Before developing systems for cellular communication, it is important to have a complete knowl-edge about how the movement of transmitter/receiver influence the first-order and second-orderstatistics of the channel dynamic parameters of that frequency band.The main difference between existing microwave communication and future mm-wave com-munication is the former system uses the omnidirectional transceiver, and the latter one will use di-rectional antennas. Due to this fact, microwave system gets access to all multipath components. So,persistence analysis of MPCs was not essential for cellular communication in microcell environ-ment. However, the mm-wave system will use directional beamforming antennas as transceivers,which will give it access to only a portion of MPCs, not all MPCs as in omnidirectional systems.At this point, in the dynamic scenario, mm-wave can work in two modes. In the first mode, it candetect available MPCs very often using beam searching protocols, which is not a feasible processas it will raise the overhead. In the second mode, it can track a particular path by adapting to itschanges with time. To design such systems, we require the analysis of persistent MPCs.41.3 ObjectivesThe primary objectives of this work are:To develop a reliable channel sounder to measure the mm-wave channel. We developed a three-stage measured data verification process to check the accuracy of our channel measurement data.We further verified our channel sounder performance using NIST channel sounder verificationsystem.To characterize multipath persistence for the mm-wave channel, including analysis of MPCspersistence at the mm-wave frequency, which was previously observed for ultra wide band (UWB),vehicle-to-vehicle (V2V), and indoor propagation channel, and even more important for mm-wavesystem design. In particular, we sought to find the distribution of the lengths of persistent MPCsand observe which environmental parameters affect the distribution parameters. The correlationbetween MPC persistence and signal strength were also analyzed. Moreover, we studied the rate ofangular change of MPCs in their lifetime and saw how this rate changes in different environments.Finally, we proposed a simulation model that can be used to incorporate finite MPC persistenceand reflected the effects into ray tracing simulated results.1.4 OutlineThe work presented in this thesis is aimed at the characterization of multipath persistence at 30GHz frequency. In this work, along with time domain analysis, we emphasize to analyze multipathcomponents (MPCs) in the angular domain, as a directional channel will require persistent MPCsin the angular domain to continue communication in dynamic scenarios.The remainder of this thesis organized is follows:Chapter 2 describes the development and operation of our vector-network-analyzer-based chan-nel sounder. We calibrated the channel sounder correctly and removed its effect from measureddata. We also verified the channel sounder with NIST verification artifact to assess its performance.We also went through three types of verification process; frequency and time domain verificationand two ray verification and confirmed that this channel sounder measured channel parameters ac-5curately. These verifications are very important to confirm that the characteristics of the channelparameters that we will identify throughout this thesis are originating due to channel effect, notfrom the channel sounder itself.Chapter 3 presents a measurement-based characterization of persistent MPCs in urban outdoorand indoor microcell environment, at 30GHz. We verified that MPC persistence fits well in log-logistic distribution that was previously used to model persistence for UWB propagation channel.It also showed how different propagation environments affect the distribution parameters. It wasfound that the existence of a weak correlation, between average persistent path power and pathlength. Moreover, this study showed that the angular variation of persistent paths are not drasticand follows a Laplacian distribution with a good consistency. Finally, we proposed a semi-Markovmodel to generate realistic persistent paths, using ray tracing simulated data and discussed themanner in which persistence will affect the performance of mm-wave systems.Finally, Chapter 4 summarizes our contributions and offers possible future directions.6Chapter 2Measurement System2.1 IntroductionNext-generation wireless communications will use mm-wave frequencies for short-range cellular,device-to-device (D2D), and other high data rate line-of-sight (LOS) applications. Before devel-oping any system for a new frequency band, having the knowledge of real propagation channelcharacteristics and accurate channel model based on channel measurement data, is important. Thisnecessitates the measurement of the propagation channel at the mm-wave frequency.The instrument which performs channel measurement using channel sounding technique iscalled a channel sounder [23]. Development of a reliable channel sounder is the first step to mea-sure the propagation channel accurately. Along with reliability, the cost is also an important factorwhich needs to be considered during the channel sounder design process. The mm-wave chan-nel sounder requires high-frequency radio frequency (RF) components such as antennas, poweramplifiers, multipliers, mixers, RF cables and other components. All of these components arevery expensive and require a large budget. We should also consider the time needed for chan-nel measurement, as channel measurement is a time-consuming procedure. The third importantfactor which needs to be taken care of is the measured signal processing complexity. After collect-ing measurement data, the next step is data reduction through which we extract required channelparameters for analysis, characterization, and modelling.7In the recent years, many mm-wave measurement campaigns have been started [22, 24–29, 33].Different groups are using different channel sounders. In [25], authors used a stepping correlatorbased channel sounder where [28] used a vector-network-analyzer (VNA) based channel sounder.Another work [29] used a multiple-input/multiple-output multiple-input/multiple-output (MIMO)channel sounder to measure mm-wave propagation channel. In this work, we measure the channelfor short-range applications, where the separation between two transmitters and receiver typicallyis not more than 100m. Again, our channel measurement is intended for pedestrian scenarios. So,we use VNA based channel sounder for channel measurement that provides us more accurate chan-nel measurement data. We further, use an omnidirectional-directional arrangement of transmitterand receiver set. On the transmitting side, we use a biconical horn antenna which is omnidirec-tional in the horizontal plane and directional in the vertical plane. In the receiving side, we use ahorn antenna, which provides enough link budget for our use cases.The primary objective of this chapter is to demonstrate that the University of British Columbia(UBC) 30 GHz channel sounder meets the requirements of persistent path measurement. For thispurpose, we develop a VNA based channel sounder and verified our measured data using threetypes of verification which confirms that our measured channel response is accurate, and the chan-nel sounder is operating as expected. We further assess the performance of the channel sounderusing National Institute of Standard and Technology (NIST) channel sounder verification artifacts.The remainder of this chapter is organized as follows. Section 2.2 discusses different types ofchannel sounders. Section 2.3 describes the configuration of our channel sounder while Section 2.4describes its operation. Section 2.5 presents the calibration and correction process and Section 2.6shows the data processing steps. Section 2.7 presents the results of channel sounder verificationwhile Section 2.8 presents the results of channel sounder performance assessment by NIST. Section2.9 concludes the chapter.82.2 ConceptIn order to perform measurement for the persistent path analysis, the designed channel soundershould meet some specifications. The designed channel sounder should be able to measure a staticchannel with an excellent dynamic range, in both indoor and outdoor environments. Further, inmm-wave measurement, it is required to measure directional information of multipath components.So, the designed channel sounder should be able to do that with a minimum measurement time.Different mm-wave channel sounders that have been used by various research groups for chan-nel measurement [22, 24–29, 33] can be broadly categorized into three groups, based on the type ofprobing signal used. They include: (i) VNA based channel sounder, which uses swept-frequencyprobing signals, (ii) correlation based channel sounder, which uses spread-spectrum probing sig-nals, and (iii) multi-carrier channel sounder, which send multiple carrier signals through the chan-nel simultaneously.Typically, for long-range outdoor measurements, correlation-based channel sounders [25], areused. In this channel sounder, a periodic pseudo-random binary sequence [31] or a periodic mul-ticarrier signal, which is known as multisine signal [32], is sent as the excitation signal. On thereceiving side, two types of architectures are generally used: (a) sliding correlator and (b) directcorrelation. A sliding correlator channel sounder first correlates the received signal with a copy ofthe transmitted one and then sample that correlated signal at or above the Nyquist rate to obtainthe channel impulse response. A direct correlation channel sounder first downconverts and digi-tizes the received signal at a high sampling rate in real time. Then it performs the correlation ofthe received signal with the ideal transmitted signal, in post-processing. Direct correlation chan-nel sounders support, very fast acquisition of channel data and offer a finer temporal resolutionfor Doppler measurements while capturing very rapid fading events. Correlation-based channelsounders use a stable 10 MHz GPS-Rubidium frequency reference to synchronize the transmitterand receiver. However, the problem of such system is that the exact propagation time of channelimpulse response is hard to extract from the measured data.9The second type of channel sounder, which usually is used for short-range indoor channelmeasurement, is the VNA-based channel sounder [28]. This channel sounder transmits a steppedfrequency sweep over the frequency range of interest and measures the frequency response of thechannel. Here, the VNA acts as both transmitter and receiver. Because it accurately measuresthe phase response of the channel, it provides accurate timing information of the channel impulseresponse. In addition to this, this channel sounder provides a high dynamic range, allowing de-tailed insight into the fading characteristics of a particular environment. However, in this channelsounder, the transmitter and receiver need to be connected with cables and cables at mm-wavessuffer from phase instability and phase nonlinearity over a wide bandwidth that can lead to dis-tortion of the frequency sweep [30]. For this reason, this channel sounder is typically used forshort-range indoor channel measurement. To extend the range, an external amplifier or a RF-over-fiber solution may be used to reduce cable loss and in the later case, it enhances the phase stabilityof the VNA-to-remote antenna link. Another problem is this system cannot measure the frequencyresponse of a signal accurately if the coherence time is less than the sweep time, i.e., it only canmeasure slowly varying or static channels.The third type of channel sounder is a multi-carrier channel sounder [29]. In this channelsounder, on the transmitting side, the vector signal generator produces a multi-carrier signal, thatconsists of multiple frequencies similar to orthogonal frequency-division multiplexing (OFDM).On the receiving side, a vector signal analyzer resolves the receiving signal into individual carrierfrequencies that were sent by the transmitter. The main advantage of this channel sounder isthat it does not require a direct cable connection between the transmitter and receiver. Instead,it uses 10 MHz GPS-Rubidium frequency references, to synchronize the transmitter and receiver.This channel sounder also does not suffer from the self-noise that limits the dynamic range ofcorrelation-based channel sounder. The main disadvantage of multicarrier signals is the mannerin which the peak to average power increases as the number of carriers increase. To some extent,this can be controlled through suitable selection of the phase relationship between the individualcarriers.10Channel sounders again can be broadly categorized into three groups, based on the arrange-ment of the transmitting and receiving antennas: (i) double-directional channel sounder, (ii)omnidirectional-directional channel sounder, and (iii) phased array MIMO channel sounder. Thedouble-directional channel sounder, used in [25, 26] mm-wave measurement campaign, uses di-rectional antennas as both transmitter and receiver. This channel sounder has only one RF chain;it rotates its transmitter and receiver using mechanical rotators to scan the whole space. This typeof channel sounding arrangement is known as virtual array system. The main advantage of thissystem is it uses directional antenna gain for channel sounding which increases the link budget.However, the major disadvantage of this channel sounder is that it requires doing an exhaustivesearch to find all available LOS and non line-of-sight (NLOS) multipath components. If the trans-mitter requires nTX steps and receiver requires nRx steps, then total nTx⇥nRx steps are needed fora full scan which requires more measurement duration to complete a measurement.The second type of channel sounder is an omnidirectional-directional channel sounder, wherean omnidirectional antenna is used on one side, and another side uses directional antenna [28].This channel sounder can measure channel in two modes, (1) receive beamforming and (2) trans-mit beamforming. In receive beamforming mode, it uses the omnidirectional transmitter to transmitsignal with an omnidirectional pattern in all directions, while the receiver is a directional antenna.To capture all available LOS and NLOS multipath, the receiver scans the whole space using a me-chanical rotator. In transmit beamforming mode, it uses a directional transmitter to transmit signalwith the directional beam towards all directions, by rotating it using a mechanical rotator, while thereceiver is an omnidirectional antenna that captures all available LOS and NLOS multipath. In thismethod, either the transmitter or receiver is rotating. While the double-directional channel sounderneeds N2 measurements, omnidirectional-directional one needs only N measurements. The maindisadvantage of this technique is, smaller link budget because of using the omnidirectional antennaon one side, so this type of channel measurement can be used for short distance measurements only.Some recent measurement campaigns [22, 29, 33] have used electronically switched beamswhich require less measurement time than the mechanical beam switching. This channel sounding11system is known as phased array MIMO channel sounder. This channel sounder use arrays ofantennas with phase shifters to form beams into different directions at both the transmitter and thereceiver. The main advantage of this channel sounding system is that it requires less measurementtime. But, phased array antennas provide large side lobes, which will require complicated signalprocessing techniques to remove antenna effect from the measurement data. Along with this,antennas of array uses different RF chains, and each RF chain requires a dedicated amplifier, phaseshifter, multiplier, mixer, and attenuator which increases the cost of the channel sounder.2.3 ConfigurationFor persistent channel analysis, we only need to measure the static channel with an excellent dy-namic range and sensitivity. We are willing to connect the transmitter and receiver with cables.This leads us towards VNA based channel sounder with fibre-optic cable. We used up and downconverters on the transmitting and the receiving side of the channel sounder, respectively to reducethe high path losses of the mm-wave signal at the cable.We originally intended to perform single polarization measurements with the vertical-verticalpolarization antenna arrangement. Later our measurement system was updated for the dual po-larization measurements. In our designed channel sounder, the antenna systems required difficultcompromises. To achieve the maximum information concerning the double-directional channel, itwas desirable to use the double-directional antenna arrangement, but that incurs long measurementtimes. However, with the omnidirectional antenna on both sides of the measurement system, wewould lose all directional information and also reduce the link budget. For the persistent measure-ment, we did an acceptable compromise by only resolving the direction of arrival and path delay.We decided to use the omnidirectional antenna on one side and directional on the other side ofthe system. With this arrangement, we also mimicked the first stage of the beam training process.Fig. 2.1 (a) shows the schematic diagram, and 2.1 (b) presents the photograph of our channel VNAbased 30 GHz channel sounder. The equipment that comprises this channel sounder is listed inTable 2.1.12(a)(b)Figure 2.1: (a) Photograph and (b) schematic diagram of our channel sounder.13On the transmitter side, a transmitting signal with frequency span 0.75 GHz to 1.75 GHz andcentre frequency 1.25 GHz was generated by the VNA. The generated signal was then passedthrough the block upconverter block-up-converter (BUC) (Norsat 7040STC-O3B-AN 4W Ka) us-ing a 400m long fibre optic cable, as shown in the Fig. 2.1 (b). The BUC upconverted the frequencyfrom 1.25 GHz to 30 GHz and provided a 55 dB gain to the output signal. Finally, the signalwith 30 GHz frequency and 36 dBm power was transmitted using a biconical antenna (SteatiteQOM-SL-26-40-K-SG-R), which presents an omnidirectional pattern in the horizontal plane andan average gain in the vertical plane of ⇡ 3.2 dB in the 29.5 - 30 GHz band. Numerical values ofall propagation parameters are given in Table 2.2.Table 2.1: List of equipment.Equipment Model No.Transmitter: Biconical antenna Steatite QOM-SL-26-40-K-SG-RReceiver: Horn antenna SAC-2309-315-S2Vector network analyzer Agilent E8362CSignal generator Agilent 8257DBlock up converter Norsat 7040STC-O3B-AN 4WMixer Sage SFB-28-N1Multiplier Sage SFA-283303213-28SF-S1Fiber optic unit Miteq SCMT-10M6G-10-20-10-E1Table 2.2: Specifications of the experiment.Parameters Numerical ValuesCenter frequency 30 GHzTransmitter Biconical, Gain: 3.2 dBi Horizontal plane:Omnidirectional, Vertical plane: DirectionalReceiver (Outdoor) Horn, Gain: 26 dBi Horizontal plane HPBW:10 Vertical plane HPBW: 13Receiver (Indoor) Horn, Gain: 23 dBi Horizontal plane HPBW:11 Vertical plane HPBW: 13Transmitter output power 32 dBmPolarization Vertical-verticalFrequency sweeping points 801Resolution bandwidth 1 kHz (Outdoor), 3 kHz (Indoor)14Table 2.3: Link budget.Parameters Numerical ValuesVNA port 1 output -10 dBmAttenuator -20.35 dBCable - 2.44 dBFiber optic unit 10 dBBias-T -3 dBBlock up converter 55 dBTransmitting antenna gain 3.2 dBiEIRP 32.1 dBmPath loss for 2 m -68 dBReceiving antenna gain 26 dBiMixer -6 dBCable -1.5 dBOn the receiver side, we used a horn antenna, a rotator, a frequency mixer, a frequency multi-plier, and a signal generator. We used two different horn antennas in our measurement campaign.In the outdoor measurement, we used a horn antenna which provides 26 dBi gain and has 10half-power beamwidth in the horizontal plane and 13 half-power beamwidth in the vertical plane.In the indoor measurement, we used a horn antenna which provides 23 dBi signal gain and has 11half-power beamwidth in the horizontal plane and 13 half-power beamwidth in the vertical plane.However, the horn antenna was mounted on an automatic rotator, which rotated it 350 inazimuth plane to collect multipath from all directions. The 30 GHz received signal was passedthrough a frequency mixer to down-convert the signal to 1.25 GHz. Here, a signal generatorequiped with a frequency doubler acted as a local oscillator. Finally, the received signal with acentre frequency of 1.25 GHz frequency was applied to the VNA. The VNA measured the scatter-ing parameter S21 from which channel gain in the frequency domain, G can be found byG= |S21|2. (2.1)Fig. 2.3 shows the measured channel frequency response of an LOS link.15Frequency [GHz]29.3 29.4 29.5 29.6 29.7 29.8 29.9 30 30.1 30.2 30.3Gain [dB]-60-55-50-45-40-35-30-25-20Figure 2.2: Uncorrected measured channel frequency response of an LOS link (transmitter-receiver separation, 2 m).To measure the propagation channel, we operated the VNA over a 1 GHz frequency span. Thetemporal resolution Ts of the measured channel frequency response is given byTs =1Fs(2.2)where Fs = BsNf and Fs = sampling frequency, Bs = frequency span, Ts = temporal resolution, andNf = number of frequency points. In our measurement system, we chose to measure the responsewith 801 points and across a frequency span 1 GHz. The frequency interval of the measurementsystem was 1.25 MHz, and temporal resolution was 0.9988 ns giving us the maximum time ofarrival of 800 ns. With this system, we were able to measure a multipath signal which was arrivingat the receiver with 80o ns maximum unambiguous delay. Further, the resolution bandwidth of thesystem was 3 kHz which provided a dwell time of 0.33 ms.16Azimuth Angle of Arrival [deg]-200 -150 -100 -50 0 50 100 150 200Gain [dB]-90-80-70-60-50-40-30Figure 2.3: Power angular profile of received signal as displayed (transmitter-receiver sepa-ration, 2m).2.4 Function and OperationTo measure the scattering parameter S21 with the VNA based channel sounder, we needed to setparameters of VNA and signal generator which was a time-consuming process. We also neededto rotate the rotator with certain steps, to steer the receiver in different directions. Standard com-mands for programmable instruments (SCPI) along with MATLAB were used to automate thesystem. SCPI commands were used to talk and instruct this measurement equipment to establishthe connection and do configuration.2.4.1 Instrument Connection and ConfigurationIn order to control the above-mentioned instruments, we used Ethernet interface and TCP/IP pro-tocols. First, we set their IP address and then we established physical connection of all equipmentas shown in the Fig. 2.1 (b).172.4.2 Measurement SetupWe used SCPI commands, to set all parameters of VNA, signal generator, and rotator. We foundsome SCPI commands from the instrument’s manual and formed some commands, following theSCPI syntax rule. Using SCPI commands, we defined the parameters, such as start frequency,stop frequency, resolution bandwidth, number of sweep points, and the reference level of VNAto configure the instrument before starting the measurement. For the rotator, we defined the startpoint, stop point and angular step size. We also specified the path and file name to store measure-ment data. VNA measured the complex voltage, S21 and stored it automatically at the predefinedlocation.2.4.3 Visualize the measurement data in MATLABWhen all 360 scan for a transmitter-receiver position was done, we plotted the power angularprofile as a function of azimuth angle-of-arrival (AOA) instantly on MATLAB to visualize the dataas shown in Fig. 2.3. This plot shows all LOS and NLOS reflected paths, coming from the nearbyscatterer and captured for transmitter-receiver separation of 2 m. We used to match these AOAwithnearby scatterer, if we had observed a satisfactory result, we proceeded to the next measurement.2.5 Calibration and CorrectionFig. 2.2 presents the measured channel frequency response for transmitter-receiver separationof 2 m. But, the measured channel frequency response, Hmeasured( f ) is a combination of actualchannel frequency response, Hchannel( f ) and the frequency response of the measurement system,Hsystem( f ). The actual channel frequency response is given byHchannel( f ) =Hmeasured( f )Hsystem( f ). (2.3)To extract the actual channel frequency response, Hchannel( f ) we need to remove the effectof measurement system, Hsystem( f ) from the measured channel frequency response, Hmeasured( f ).For this purpose, we first need to characterize the imperfections of the measurement system, by18Figure 2.4: Photograph of system response, Hb2b( f ) measurement.measuring Hsystem( f ). The measurement of frequency/impulse response of measurement system isknown as calibration procedure. We performed calibration by connecting transmitting and receiv-ing ends, back to back without using any antenna. Fig. 2.4 presents the photograph of one of ourcalibration process where we are measuring back to back system response, Hb2b( f ). The antennaand the channel were replaced with a phase stable coaxial cable and two 30 dB attenuators. Weused two 30 dB attenuators to attenuate the transmitting signal by 60 dB, so that, high receivingsignal does not damage any RF equipment or does not distort the signal by generating nonlinearfrequency components.Fig. 2.5 (a) and Fig. 2.5 (b), show the frequency and the phase response of the measurementsystem, respectively. The frequency response of the measurement system exhibits a 5 dB variation.The phase response is linear. However, during this system calibration process, we used an RF cableand two attenuators to replace over-the-air (OTA) path, but these three RF components were not19Frequency [GHz]29.3 29.4 29.5 29.6 29.7 29.8 29.9 30 30.1 30.2 30.3Gain [dB]-80-75-70-65-60-55-50-45-40(a)Frequency [rad/s] ×10111.84 1.85 1.86 1.87 1.88 1.89 1.9 1.91Phase [rad]-1800-1600-1400-1200-1000-800-600-400-2000Slope = -2.7e-07(b)Figure 2.5: (a) Frequency and (b) phase response of back-to-back system response, Hb2b( f ).20Frequency [GHz]29.3 29.4 29.5 29.6 29.7 29.8 29.9 30 30.1 30.2 30.3Gain [dB]-6.1-6-5.9-5.8-5.7-5.6-5.5-5.4(a)Frequency [rad/s] ×10111.84 1.85 1.86 1.87 1.88 1.89 1.9 1.91Phase [rad]-30-25-20-15-10-505(b)Figure 2.6: (a) Frequency and (b) phase response of the RF cable.21Frequency [GHz]29.3 29.4 29.5 29.6 29.7 29.8 29.9 30 30.1 30.2 30.3Gain [dB]-30.7-30.6-30.5-30.4-30.3-30.2-30.1-30(a)Frequency [rad/s] ×10111.84 1.85 1.86 1.87 1.88 1.89 1.9 1.91Phase [rad]-0.0500.050.10.150.20.250.30.350.40.45(b)Figure 2.7: (a) Frequency and (b) phase response of the 30 dB attenuator-1.22present in actual channel measurement, so we have to discard the effect of these RF components.To do that, we characterized high-frequency RF cables and attenuators separately.Fig. 2.6 (a) and Fig. 2.6 (b), show the frequency and phase response of the RF cable, respec-tively. Fig. 2.6 (a), shows a very flat frequency response with a power fluctuation of 0.8 dB. Thephase of the RF cable is also very linear as shown in Fig. 2.6 (b). Fig. 2.7 shows frequency andphase response of the first attenuator. From these figures, it is observed that the attenuator and theRF cables show a linear phase response and flat gain, as expected. The second attenuator shows asimilar frequency and phase response as the first attenuator.The final system response also includes the effect of transmitting and receiving antennas, in theform of antenna gain, which were not present in the back-to-back system response measurement.Thus the system response is given byHsystem( f ) =Hb2b( f )⇥GTX ⇥GRXHcable( f )⇥Hattenuator1( f )⇥Hattenuator2( f ) (2.4)where Hsystem( f )= complex frequency response of the measurement system, Hb2b( f )= back-to-back response of the measurement system, GTX = gain of the transmitting antenna, GRX = gainof the receiving antenna, Hcable( f ) = complex frequency response of RF cable, Hattenuator1(w) =complex frequency response of the attenuator-1, Hattenuator2( f ) = complex frequency response ofthe attenuator-2.Figure 2.8 represents the channel frequency and phase response for 2 m separation betweenthe transmitter and receiver, after removing the complex frequency response of the system. Thisfrequency response shows a 7 dB fluctuation. From Fig. 2.8 (b), we also observe that the phaseresponse of the link is linear although there are some small fluctuations.2.6 Data ProcessingAs discussed in Section 2.3.1, we measured our channel in the frequency domain using a VNA-based channel sounder, but we analyzed the results in the time domain. In order to convert fre-quency domain data to the time domain, we used the inverse fast Fourier transform (IFFT). Before23Frequency [GHz]29.3 29.4 29.5 29.6 29.7 29.8 29.9 30 30.1 30.2 30.3Gain [dB]-90-85-80-75-70-65-60-55-50-45-40(a)Frequency [rad/s] ×10111.84 1.85 1.86 1.87 1.88 1.89 1.9 1.91Phase [rad]-40-35-30-25-20-15-10-50(b)Figure 2.8: (a) Frequency and (b) phase response of an LOS link after removing the effect ofthe measurement system (transmitter-receiver separation, 2 m).24Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Channel Impulse Response [dB]-140-130-120-110-100-90-80-70-60Hanning WindowHamming WindowKaiser WindowFigure 2.9: Channel impulse response of an LOS link with different window functions.IFFT, we apply a window function to the measured data where measured frequency response ismultiplied by frequency domain window function. In this section, we discuss how we configuredthe windowing and IFFT operation in our MATLAB scripts.2.6.1 WindowingUse of finite frequency span in IFFT introduce artifacts in the converted data. To remove thisartifact from data, a window function is commonly used before doing IFFT. Window functioncuts the edges of the signal which reduces the effect of artificial transients produced due to finitefrequency length [34]. The complex impulse response resulting from applying window is given byh(t)unscaled = F1(Hchannel( f )Hwindow( f )), (2.5)where Hchannel( f ) is the corrected channel frequency response, Hwindow( f ) is the response of thewindow in the frequency domain, and F1 denotes the inverse Fourier transform.25Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Channel Impulse Response [dB]-115-110-105-100-95-90-85-80-75-70-65Without windowingWith windowing (Hanning)Figure 2.10: Channel impulse response of an LOS link showing the effect of windowing.Different types of window functions are available in the literature, each of them has their owncharacteristics. Hanning, Hamming, and Kaiser are three window functions are often used beforeIFFT. In Fig. 2.9, a result is showing the performance comparison of these three window functions,and we observe that the results are very similar. Hence, we decided to use Hanning window [34]for our measurement data processing.Fig. 2.10 compares the impulse response of an LOS link before and after using Hanningwindow from which we see that windowing has reduced the side lobes of the signal. However,windowing also reduces the total energy content of the frequency response. To compensate for thisdifferent types of scaling are used, including: 1) maximum scaling, 2) area scaling, and 3) discretesum scaling.The maximum scaling scales the channel impulse response so that the main peak has a magni-tude of one.h(t)maxscaling =h(t)unscaledmax|h(t)unscaled| , (2.6)26where h(t)unscaled is the channel impulse response before scaling process.The area scaling scales the channel impulse response byh(t)areascaling = h(t)unscaledpNf d fq(R Nff=1 |Hfilter( f )2|d f ), (2.7)where Hwindow( f ) is the response of the window in the frequency domain and Nf is the number offrequency points.The discrete sum scaling scales the channel impulse response by the discrete sum, given byh(t)discretesumscaling = h(t)unscaledpNfq(ÂNff=1 |Hfilter( f )2|), (2.8)where Hwindow( f ) is the response of the window in the frequency domain and Nf is the numberof frequency points. In our data processing, we used discrete sum scaling to compensate for theenergy lost due to windowing. Fig. 2.11 shows the channel impulse response before and afterapplying scaling. From this figure, we observe that the scaling has shifted the channel impulseresponse by 4.2 dB.2.6.2 Inverse Fast Fourier Transform (IFFT)IFFT converts a frequency domain function into the time domain. In this work, we used aMATLAB built-in function to perform IFFT which computes the inverse discrete Fourier trans-form (IDFT) of the frequency domain data using a fast Fourier transform algorithm. The complexchannel impulse response resulting from applying IFFT is given byh(t)unscaled = ifft(Hchannel(f )Hwindow(f ),symflag) (2.9)where Hchannel( f ) is the corrected channel response, Hwindow( f ) is the window response in thefrequency domain, ifft denotes the inverse Fourier transform, and symflag is the data symmetrytype. In this IFFT operation, input data can be treated in two ways: 1) Symmetric and 2) Non-27Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Channel Impulse Response [dB]-115-110-105-100-95-90-85-80-75-70-65Without scalingWith scalingFigure 2.11: Channel impulse response (CIR) of an LOS link showing the effect of scaling.symmetric. The symmetric option for the IFFT is applicable only when the input data is real-valued.In our measurement campaign, we measured the complex frequency response of the channel,so it is not confirmed that data are conjugate symmetric. Thus, we used nonsymmetric option forthe IFFT to convert our frequency domain data to the time domain. Fig. 2.12, compares the outputof IFFT, assuming symmetry and assuming nonsymmetry. We see that nonsymmetry assumptionof data provides a better result. In our data processing during IFFT operation, we used finitefrequency length. The length of IFFT was determined automatically by the MATLAB function tofit the given data.Fig. 2.13 (a) shows the channel impulse response of an LOS link for transmitter-receiverseparation of 2 m after doing all signal processing.28Time of Arrival [ns]0 50 100 150 200 250 300 350 400Channel Impulse Response [dB]-140-130-120-110-100-90-80-70-60-50-40Symmetric IFFTNon symmetric IFFTFigure 2.12: Channel impulse response of an LOS link showing the effect of symmetric andnon-symmetric IFFT.2.7 VerificationIn this section, we discuss the verification process that we have used to verify the accuracy of ourcollected data and also the assess the performance of our channel sounder. We have conductedthree types of verification, (i) time domain verification, (ii) frequency domain verification, and(iii) two ray verification.2.7.1 Verification in Time Domain of LOS PathThe goal of this verification is to verify that the timing delay of LOS signal agrees with the freespace path delay and the path gain of the signal agrees with theoretical path gain. For this veri-fication, first, we placed our transmitter and receiver 2 m apart and measured the LOS signal bypointing the receiver towards the transmitter. Then we calculated the time-of-arrival (TOA), t of29Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Channel Impulse Response [dB]-115-110-105-100-95-90-85-80-75-70-65TOA = 6 nsPath gain = -68.18 dB(a)Transmitter-Receiver Separation [log scale]0 2 4 6 8 10 12 14 16 Gain [dB]-100-95-90-85-80-75-70-65Measured path gainTheoretical path gainRegression line (theoretical path gain)Regression line (measured path gain)(b)Figure 2.13: (a) Channel impulse response of an LOS link (transmitter-receiver separation,2m) (b) Theoretical vs. actual path gain plot for LOS links (transmitter-receiver sepa-ration, 2-16 m, with 2m steps).30LOS signal using the following equationt = dc, (2.10)where c = speed of light and d is the physical separation between the transmitter and receiver.According to Eqn. 2.10, the calculated TOA, t for 2 m of signal travel is 6.67 ns. We comparedthis calculated TOA with the measured TOA. From Fig. 2.13 (a), which presents channel impulseresponse of that LOS link, for transmitter-receiver separation 2 m, we observe that the TOA ofLOS component is 6 ns, which is a good match with the result we calculated from the physicalseparation.Tomatch the free space path gain of the same LOS signal with the theoretical one, we calculatedthe free space path loss for the similar transmitter-receiver separation using the Friis transmissionformula [35], as the path loss is the inverse of path gain. Friis formula is given byFSPL= (4pdl)2 (2.11)where FSPL is free space path loss, d is the separation between the transmitter and receiver, l is thewavelength of the transmitted signal. According to the Friis equation, for a 2 m transmitter-receiverseparation and 30 GHz transmitted signal, the path gain is -68.008 dB and from Fig. 2.13 (a), wefound the path gain of the same measured LOS link is -68.18 dB, which is a good match. To see theconsistency of this result, we calculated path gain of a series of receiver positions and comparedthe measured path gain of the same links; results are shown in Fig. 2.13 (b). From the result,we observe that for the first three receiver positions theoretical results match with the measuredpath gain with a good consistency. However, from receiver four, we observe some deviation ofmeasured path gain from the theoretical one. This may happen due to the fading effect of LOS linkby the ground reflections because for longer distances ground and building reflections fall withinthe beamwidth of the antenna.312.7.2 Verification in Frequency Domain of LOS PathIn order to ensure the accuracy of the measurement system, we also verified our measured data inthe frequency domain. In this verification, we calculate the absolute separation between transmitterand receiver from the channel phase response and match that with measured transmitter-receiverseparation. For this purpose, we derive an equation by using Fourier transform and the shiftingproperty of time-delayed Dirac delta function. The Fourier transform of time-delayed Dirac deltafunction can be written asH( jw) =Z ••d (t t0)e jwtdt, (2.12)H( jw) = e jwt0 = e jt0w = e jf . (2.13)This shows that the Fourier transform of shifted impulse is a complex exponential. Using thisequation, the phase response of the signal, f can be written asf =t0w. (2.14)Equation 2.14 (a) shows that phase of the signal is a function of w and t0 is a constant. We measurethe phase response of LOS signal by placing transmitter and the receiver 2-m apart and pointingthe receiver towards the transmitter. We plot the phase of the measured signal as a function offrequency in Fig. 2.14 (a), where we see that the slope of the phase response is t0. This slopet0 represents the TOA of the signal which is 5.1 ns. When we calculate physical separation oftransmitter-receiver using Eqn. 2.10, we get 1.53 m, where the actual physical separation of thetransmitter and receiver was 2 m.To observe the consistency of the result, we measured the phase response for eight transmitter-receiver locations, first by placing the transmitter and the receiver at 2 m apart and then increasingthe separation by 2 m step. We calculated the transmitter-receiver separation from the measuredphase response using the method mentioned above and compared them with actual separation. The32Frequency [rad/s] ×10111.84 1.85 1.86 1.87 1.88 1.89 1.9 1.91Phase [rad]-40-35-30-25-20-15-10-50Phase response Regression lineSlope = -5.1e-09(a)Actual Transmitter-Receiver Separation [m] 0 2 4 6 8 10 12 14 16 18 20Estimated Transmitter-Receiver Separation [m]02468101214161820Transmitter-receiver separationRegression lineSlope = 1Standard deviation = 0.048(b)Figure 2.14: (a) Estimation of the transmitter-receiver separation from phase response and(b) Estimated vs. actual transmitter-receiver separation plot.33results are presented in Fig. 2.14 (b). From the figure, we observe that there is a linear relationbetween estimated and actual transmitter-receiver separation. We further observe that the slope ofthe regression line of the scattered plot is one and the standard deviation of scattered points fromthat line is 0.048, which confirms the consistency of the result.2.7.3 Two Ray VerificationIn the third stage, we verified our data using a two-ray measurement. In this verification first,we measured a NLOS reflected path by placing our transmitter and receiver 1.5 m apart from areflector. The separation between the transmitter and receiver was 2 m. The measurement scenariois presented in Fig. 2.15. We calculated the expected TOA and angle-of-arrival (AOA) of thisNLOS path, based on the physical separation of reflector, transmitter and the receiver and matchedthat with the measured results.Figure 2.15: Two-ray measurement scenario.In the measurement, the transmitter-receiver route was 1.5 m away from the reflector wallwhich provides the distance travelled by the reflected path, dNLOS = 3.66 m. When we calculatedthe TOA for this distance, dNLOS using Eqn. 2.10, we found it to be 12.2 ns. Fig 2.16 showschannel impulse response of that measured reflected path which also shows a delay of 12 ns. Thisreflected path arrive at the receiver at azimuth angle 57. When we calculated the AOA of thissignal using Snell’s law [36] we also observed same AOA for this path. This verifies that AOA and34Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Channel Impulse Response [dB]-125-120-115-110-105-100-95-90-85-80TOA = 12 nsFigure 2.16: Channel impulse response of a NLOS reflected path.TOA of the measured signal are correct. Fig. 2.3 shows the power angular profile for 360 scan ofthis measurement arrangement. We also observe that the AOA of the reflected path is 57.2.8 Channel Sounder Verification by NISTThe NIST has formed the 5G mmWave Channel Model Alliance to coordinate channel measure-ments and modelling research at mm-wave frequencies for 5G communications. In order to allowalliance participants to confidently utilize data from different measurement groups but nominallysimilar environments to develop channel models, NIST has developed a channel-sounder verifi-cation artifact which will assess the performance of different channel sounders used by differentchannel measurement groups [37]. The artifact contains splitters, combiners and coiled cableshoused in a temperature-controlled housing. This verification artifact can emulate a channel bygenerating a signal with a known delay and path loss with which it is possible to check how accu-rately a channel sounder can measure signal parameters.35Figure 2.17: Setup of channel verification process using NIST channel sounder verificationartifacts.To conduct a channel sounder verification campaign, NIST came twice to UBC Radio ScienceLab, on 24/25 October 2016 and 19May 2017. The key outcomes of these campaigns are threefold.Firstly, they recommended some minor improvement in our channel sounder and gave suggestionsabout the best practices for the data post-processing. Secondly, they verified the accuracy and thedynamic range of the channel sounder. Thirdly, they shared their experience about propagationchannel measurement and analysis.One of the typical verification measurement of that verification campaign is presented in Fig.2.17 and the results are shown in Fig. 2.18. In this verification process, two multipath componentswere generated using the channel-sounder verification artifact and measured using our 30 GHzVNA based channel sounder. From Fig. 2.18, we observe that our channel sounder was able to360 100 200 300 400 500 600 700 800 900Time of Arrival [ns]-160-150-140-130-120-110-100-90-80-70-60Gain [dB]TOA = 54 ns Gain = - 78.7dBTOA = 10 ns Gain = - 70.54 dBDynamic range = 45 dBFigure 2.18: A typical verification result from the NIST channel sounder verification cam-paign.accurately measure the gain and TOA of these two multipath components and achieve a dynamicrange of 45 dB.UBCwill continue to work with NIST to analysis uncertainty of mm-wave propagation channelmeasurement [38], [39] which will be incorporated into the microwave uncertainty framework.This uncertainty analysis will include the uncertainties due to antenna misalignment which wasalso observed in our measurement campaign. We will further work together to extend their channelsounder measurement.2.9 SummaryIn this chapter, we described the 30 GHz, VNA based omnidirectional-directional channel sounderthat we used to generate the results presented in Chapter 3. By using an omnidirectional transmit-ting antenna, we were able to cut the measurement time compared to double-directional channel37sounder. We have also spent less amount than MIMO channel sounder. Our primary goal was tomeasure 30 GHz propagation channel for D2D and other short-range applications to analyze per-sistent multipath components. Using the VNA based channel sounder we were able to measure thechannel with an excellent dynamic range. We also went through three types of verification process;time and frequency domain verification and two ray verification, and confirmed that this channelsounder measured multipath parameters, such as path gain, TOA, and AOA, accurately. Theseverifications are very important to confirm that the characteristics of the channel parameters thatwe will identify throughout this thesis are originating due to channel effect, not from the channelsounder itself. To check the reliability of the channel sounder we assess its performance with NISTverification artifact. This verification artifact generates multipath components at a predefined delayand reduced power. We matched our measured data with that predefined delay and received powerand found it to be matched.38Chapter 3Characterization of Multipath Persistence3.1 IntroductionAs a mobile terminal traverses the coverage area of a transmitter, it moves concerning the scatterersand reflectors. As a result, individual multipath component (MPC) will evolve in characteristicways. The manner in which MPCs appear, change and disappear, also referred to as the birth-death process and has been studied in both indoor and street canyon environments in various bands,allocated to personal communications.When anMPC evolves in a dynamic scenario, it persists in the propagation channel by changingits traceable features, such as azimuth angle-of-arrival (AOA), time-of-arrival (TOA), and receivedpower. MPC may also drop temporarily, i.e., switch from an active to an inactive state. Theduration of the active state of an MPC is referred to as the persistence, where the duration of theinactive state is defined as the dropout. However, the dropout phenomenon may happen due toseveral reasons, one of them is blockage of the path by an intervening obstacle, which may occurbecause of the limited ability of mm-wave signals to diffract around the obstacle. MPCs can alsodisappear temporarily due to the presence of a physical gap in the reflecting surface. Further,reduction of received power of MPCs may happen due to the patch of the surface that correspondsto the reflection point having poor reflecting properties.39Such dropouts may have significant effects on the performance of the mm-wave system, in-cluding capacity/throughput and bit error rate (BER). While the link is being established, withthe increase in the time device needed to perform beam tracking. However, dropouts reduce thereliability of the beam tracking process. This reduction in the reliability of beam tracking fur-ther decreases the capacity of the communication link and also increases BER. To develop themitigation strategies for dropouts, the knowledge of the statistics of link persistence, in dynamicscenarios is required.Persistence results due to the appearance and disappearance of MPCs in the dynamic scenario.Study of this MPCs appearance and disappearance are not new for wireless communications. MPCevolution had previously been reported for indoor wideband spread spectrum applications [40, 41],where authors modelled appearance and disappearance of MPC based on deterministic ray tracingresults. Some works have also studied evolution of MPCs in indoor scenarios for space divisionmultiple access (SDMA) and multiple-input/multiple-output (MIMO) applications [42–45], basedon measurement data. Along with modelling MPC’s evolution, these works also modelled thespatio temporal variation of paths within their lifespans. Evolution of MPCs further has beenanalyzed for ultra-wide-band (UWB) localization applications [45]. In addition to this, few studieshave also been reported which were focused on modelling and characterization of dynamic MPCsparameters for vehicle-to-vehicle (V2V) communications [46] – [47].In mm-wave beam tracking, analysis of dynamic nature of MPC persistence is even more im-portant, because the mm-wave system will use highly directional antennas. This will make mm-wave connections intermittent [49] because directional antenna will have access only to thoseMPCs of the propagation channel, which are impinging in the beamwidth of the antenna; hencethis links will require more adaptation with the movement of the transmitter/receiver. It has alsobeen observed that the signal propagation in mm-wave frequency is dominated by line-of-sight(LOS) and lower-ordered reflected paths [52]. Further, in high frequency, the scattered compo-nents will provide the weak signal. Because of this nature, the mm-wave channel will have sparseMPCs compared to the lower frequencies. This will make mm-wave channel very different than40existing lower frequency channel, and the models of persistent paths mentioned above, which weredesigned for lower frequency applications, are not valid for mm-wave channel analysis.In the recent years, first-order statistics (path loss, delay spread, angular spread and K-factor)of mm-wave channels in indoor and outdoor environments have extensively been studied [11–20].On the other hand, relatively few authors, e.g., have previously sought to characterize second orderparameters, such as autocorrelation length [21], Doppler spread [22]. However, persistence whichis another important second-order characteristic of mm-wave channel still needs to be studied.Along with dynamic channel model, deterministic ray tracing simulation technique also con-sidered as an effective means for the dynamic spatio-temporal analysis of MPCs for mm-wavefrequencies. However, site-specific ray tracing simulations are not enough to capture all persis-tent paths characteristics very precisely. Simulation can only capture macro-persistent propagationpaths. Macro-persistent paths originate due to the simple variation of the reflectors. However,complex wall structures, including metal-framed glass wall, inclined wall, windows, beams, gen-erates micro-persistent paths which are difficult to capture with ray tracing simulation (micro andmacro persistence are discussed in Section 3.2). A very detailed database of building outline isrequired for the accuracy of simulated results, which may be impractical. Also, the number ofMPCs may be comparatively less than the actual environments. This shortcoming of ray tracingsimulation software has motivated us to characterize MPC persistence for dynamic scenarios basedon measurement data.To the best of our knowledge, ours is the first study to characterize multipath persistence. Theobjectives of this chapter are threefold. Firstly, to confirm the existence and also the large andsmall scale nature of persistence MPCs in the mm-wave channel. For this purpose, we carried outan extensive measurement campaign and measured propagation channel for indoor and outdoormicrocell scenarios in the University of British Columbia (UBC), Vancouver, at 30 GHz. We alsointroduce the concept of micro and macro persistence and show, how they explain small and large-scale behaviour of the mm-wave propagation channel. We further develop an algorithm to trackpersistent paths and also define appropriate path tracking parameters for mm-wave frequencies.41Our second objective is to characterize persistence statistics based on measurement data anddetermine the extent to which our knowledge and understanding of MPC persistence at lowerfrequencies apply at mm-wave frequencies. We develop a statistical model for the rate of AOAchange of persistent paths. Moreover, we determine how distribution parameters of multipathpersistence and rate of AOA change are dependent on different propagation channel. We furtherobserve the correlation between average received power of persistent path and path length.Our third objective is to identify common behaviours of persistent MPCs and propose a sim-ulation model that can be used to introduce finite persistence into ray tracing simulations basedupon building outlines that do not contains all details. We further provide a thorough discussion ofhow persistence will impact system performance, such as throughput/capacity, beam training, andbeam tracking schemes.The remainder of the chapter is organized as follows. Section 3.2 explains the physical basisof multipath persistence and our persistent path tracking algorithm. Section 3.3 describes ourmeasurement system, measurement scenarios, and data acquisition process. Section 3.4 analysesthe results. In Section 3.5 we propose a simulation model that can be used to make MPCs that isgenerated by deterministic ray tracing, more realistic. Section 3.6 describes how persistence willimpact mm-wave system performance and Section 3.7 concludes the chapter.3.2 Concept3.2.1 Physical Basis of PersistenceIn this section, the physical basis behind the formation of persistent paths will be explained. Thepersistence or lifetime of an MPC refers to the period over which its traceable features exhibitsdifferential changes with differential movement of the receiver, within the period between birthand death of an MPC. Persistence may result due to two facts, (i) geometrical relationship betweenthe reflecting surfaces and MPCs and (ii) reflectors handoff.In order to explain the physical basis of persistent paths, we use a simple reflector with a smallirregularity and place six receivers from position rx1 to rx6, as shown in Fig. 3.1. The MPC,42Figure 3.1: Physical basis of persistent paths.which is originated by the reflection of the wall is captured from these six receiver locations. Fromreceiver position rx1 to rx3 MPC changes its signal strength, TOA and AOA differentially. Atreceiver position rx4, MPC does not arrive at the similar way and appears like a dropout for thisreceiver. The small irregularity in the reflecting surface causes this dropout. However, reflectedsignal drops briefly and appears again at receiver position rx5 due to the end of that irregularity.After seeing the effect of this temporary dropouts, we decide to categorize persistence in twocategories, (i) macro persistence and (ii) micro persistence. In large-scale persistence analysis, thistypes of irregularities do not affect the path considerably so, the whole path appears as a singlepersistent path of length L and we define such path as a macro persistent path. On the other hand,in the small scale persistence analysis, even the presence of small irregularities in the reflectingsurface introduce temporary dropouts, as shown in Fig. 3.1, and the whole path appears as twopersistent paths of length, l1 and l2. We define such paths as micro persistent paths.Second types of the persistent path may result due to the similarity between two neighbour-ing MPCs which have originated by different reflectors, we define this phenomenon as reflectors43handoff. These persistent paths sometimes show an abrupt change of one of three persistent pathtracking parameters.3.2.1.1 Effect of street width on the rate of AOA changeTo establish a communication link receiver generally choose a path that has highest signal strength.The receiver receives the strongest MPC, regarding received signal strength, from the nearby re-flector. Thus the nearby reflector has an important role. When receiver traverses, this MPC changesits AOA along with other parameters, such as TOA and received power.To track this MPC with narrow beamwidth antenna of a mm-wave system, it is important toknow how fast AOA of this MPC is changing. When the separation between the reflector andreceiver route is large, AOA of MPC changes moderately, as shown in Fig. 3.2 (a). However, whenthe separation between the reflector and receiver route is small, AOA of MPC changes at a fasterrate, as shown in Fig. 3.2 (b). Which leads to the conclusion, the rate of AOA change of MPC isinversely related to the separation between nearby reflector and transmitter/receiver route.3.2.2 Tracking and Modelling ApproachesThis subsection discusses the tracking and modelling approaches of the persistent path that havebeen adopted previously for other applications.3.2.2.1 Tracking persistent pathPersistence effect is not very common in the macrocellular environment where most paths are theresult of multiple diffraction and reflections. However, in the microcellular environment, NLOSpropagation is highly dominated by specular components which arise the importance of persis-tence in microcellular communication. Previously persistence or the lifetime of MPCs has beencovered in three different applications; (i) indoor scenarios, (ii) V2V communications, and (iii)UWB localization purposes.In [42], authors analyzed the spatio-temporal variations of paths within their lifespans for anindoor environment. In order to detect a persistent path authors used linear least squares regression44(a)(b)Figure 3.2: AOA change for (a) large and (b) small separation between main reflector andtransmitter-receiver route.45to find the best line fit through all data points of each path in the form of y = kx+ b. Based ontheir findings authors assumed that direction of change of MPCs on the spatio-temporal domainis independent of the location thus they determined gradient k of each straight line by a randomvariable.On the other hand, in [48] authors present a low-frequency temporal tracking of cluster cen-troids for MIMO channel data. To identify the time variant behaviour of a cluster, this work trackedcluster centroids over several time window. This cluster tracking algorithm has chosen a multipathcomponent distance (MCD) between cluster centroids as a distance metric and track cluster cen-troids by minimizing this MCD. However, this work was intended for MIMO application, and theydid not consider the power variation of the cluster and characterized each path by its TOA andAOA.In UWB localization, the persistent path was used to detect mobile devices in the absence ofdirect paths [45]. A persistent path was tracked by comparing TOA and AOA of MPCs, of theconsecutive receiver positions. In this method, the current paths choose the next consecutive pathsthat have the path metrics (TOA+AOA) closest to the metrics obtained. Similar to other indoorwireless communication applications, this work also considered TOA and AOA of MPCs as pathtracking parameter as these two parameters are important to track a path which is originating fromthe same reflector. Authors did not consider received power of the MPC as path tracking parameterbecause for localization applications power variation consideration is not so important.In V2V scenarios, the lifetime of MPCs was analyzed in the large-scale macrocellular environ-ments [46]. To understand the dynamic process of the vehicular radio channel, this work trackedthe temporal behaviour of individual MPCs by calculating the distance between two consecutiveMPCs regarding path gain and TOA. This work was intended for low frequency (5.7 GHz) whereomnidirectional antennas are used as the transceiver, so authors did not consider AOA as MPCtracking parameter.Along with these above-mentioned works, ray tracing simulation method also has been adoptedwidely for dynamic propagation channel analysis. Ray tracing simulation gives us insight about46how MPCs changes its parameters with the movement of the transmitter/receiver. However, due tolower complexity of the ray tracing database, it shows a very smooth variation of these parameterswith travelling distance. However, in the actual scenario, because of the complexity of the buildingarchitecture, the smooth variation of MPC parameters are being interrupted. Ray tracing presentsthe evolution of MPCs as deterministic components. However, the complexity of the infrastructureis not possible to capture using deterministic model.3.2.2.2 Modelling persistent pathAlthough persistence of MPCs has never been studied for mm-wave but has been studied at lowerfrequencies for applications such as indoor [42], localization [45] and V2V [50]. For indoor ap-plications, lifetime of MPCs was modelled using an exponential distribution [41], [42]. The expo-nential probability distribution function (PDF) can be written asf (x;l ) = lelx, (3.1)where l is the rate parameter. On the other hand Ferit et al. [45] proposed to use log-logisticdistribution to model persistence. Log-logistic PDF can be expressed asf (x;µ,s) = 1sxez(1+ ez)2;x> 0 (3.2)andz=log(x)µs, (3.3)where µ and s are the scale and shape parameter, respectively. The reason behind using log-logistic model was, this model can describe longer persistent path that is produced by longer walls.The Exponential models decay too quickly to describe this behaviour.Multipath persistence has also been analyzed for V2V applications. He et al. [50], used atruncated Gaussian distribution to model lifetime/persistence of MPCs. The truncated Gaussian47distribution for a< x< b can be written asf (x;µ,s ,a,b) =f( xµs )s(F(bµs )F(aµs )), (3.4)where F and f denote the cumulative distribution function and probability density function ofthe normal distribution, respectively, µ is the location and s > 0 is the scale parameter of thedistribution, a and b are the minimum and maximum support of the distribution. However, excepta< x< b, truncated Gaussian PDF will be zero.3.2.2.3 Time domain dynamic path analysisFew studies have been reported which were focused on modelling and characterization of dy-namic propagation path parameters for outdoor scenarios [52] –[53]. In [52], authors used a quasi-deterministic approach to model propagation path parameters at 60 GHz frequency band. On theother hand, Santos et al. [53], used a geometry-based approach to characterize propagation pathsfor the ultra-wideband channel. In the above-mentioned works, propagation paths were trackedand described only in the time domain.Fig. 3.3 (a), shows a typical time domain representation of dynamic propagation path, wherethere are some LOS and NLOS paths with different length. But, for mm-wave propagation pathanalysis, along with time domain, knowledge of angular domain variation of propagation paths arealso important which is not available in time domain analysis.Fig. 3.3 (b), presents three reflected paths. These three paths are located inside a single ellipse.Although these paths are arriving at the receiver in three different angles, their TOA is same. So,in time domain analysis all of these three, will appear in a single line and persistence and dropoutsof one this path will not be reflected in this time domain representations. For omnidirectionaltransceiver blockage/dropout of one of these paths will not affect the persistence of the link astwo other paths are still available to the transceiver. However, this not similar to the directionaltransceiver. A directional antenna will only have access to one of these paths at a time so, dropoutof that path will affect the persistence of that link. This observation provides us with an intu-485 10 15 20 25 30 35Transmitter-Receiver Separation [m]050100150200250300350Time of Arrival [ns]Full length NLOS persistent pathsLOS path(a)(b)Figure 3.3: Time domain analysis of propagation paths (a) persistent paths with moving re-ceiver and (b) physical basis.49ition that in the mm-wave dynamic propagation path analysis needs to be presented in space-timerepresentation. Thus, time domain representation does not provide the complete knowledge.3.2.3 Persistent Path Tracking AlgorithmIn order to track persistent paths, we calculate the distance between two consecutive MPCs re-garding their TOA, AOA and received power and connect MPCs which has the minimum distance.This distance was previously defined as MPC distance (MCD) and is used to track a time-varyingcluster [48]. In our algorithm, we calculate this distance using Euclidean distance metric.We denote the ith MPC received at the jth receiver byXi( j) = {ti( j),fi( j),Pri( j)}, (3.5)where ti( j), fi( j) and Pri( j) represents TOA, AOA and received power of ith MPC correspondingto jth receiver respectively and i = 1 . . .Lj, Lj is the total number of MPCs captured at receiverlocation, j. When receiver changes its position, some paths continue with the differential changein their TOA, AOA and received power and tracked as persistent paths, some dies and some newpaths evolve. The tracked persistent paths, Tk( j) at receiver location j is defined asTk( j) = {tk( j),fk( j),Prk( j)}, (3.6)where tk( j), fk( j) and Prk( j) represents TOA, AOA and received power of kth tracked MPC. Atreceiver position j, to find if MPC Xj,i belongs to a persistent path we calculate the distance of thisMPC with all available persistent paths, Tk( j 1) of receiver position j 1 using the followingEuclidean distance metricdk,i =q(tk( j1) ti( j))2+(fk( j1)fi( j))2+(Prk( j1)Pri( j))2 (3.7)50where, k= 1 . . .Lj1. After that, we connect this MPC with a persistent path, Tk( j1) and updateit as Tk0( j) as followsTk0( j) =8><>:Xi( j), min(dk,i) Thdist0, min(dk,i)> Thdist (3.8)where Thdist is the distance threshold. This distance threshold depends on the fact that how muchfluctuation of MPC parameters, the system can tolerate.For the first receiver location, there are no previous persistent paths, with which MPCs of thatposition can be connected so, we assume Tk( j1) is to be zero at this position.3.2.3.1 Estimating New PathsAfter connecting all MPCs of receiver position j with position j 1, some MPCs remains whichcould not get connected, these MPCs are newborn MPCs of receiver position j. This newbornMPCs will be added in Tk0( j) and form Tk( j) at receiver position j.3.3 Measurement Campaign3.3.1 Measurement SystemThe measurement platform used for this study is a vector-network-analyzer (VNA) based channelsounder, which we described in Chapter 2. We use a 30 GHz channel sounder with 1 GHz band-width and 1ns temporal resolution to characterize the propagation channel. The VNA operates at1.2 GHz centre frequency with a 1.2 GHz/30 GHz up and down converters. An omnidirectional bi-conical antenna with 3 dBi gain is used as the transmitter, which transmits 32 dBm radio frequencysignal. As for the receiver, we use a highly directional, vertically polarized horn antenna with 23dBi gain and 11 half power beamwidth. Also, an automatic rotator is used to rotate the directionalreceiving antenna along 360 in azimuth plane to emulate a beam steering receiving sensor to col-lect MPCs from all directions. For the sake of simplicity, MPCs are measured from the azimuthplanes only. To verify the accuracy of our channel sounder, we calibrate our measurement systemproperly. For calibration, we perform a reference measurement by connecting transmitting and51receiving ends, back to back without using an antenna with a phase stable RF cable and two 30dB attenuators. Both, the cable and attenuators are characterized separately, and their effects areremoved afterwards.3.3.2 Measurement ScenariosIn order to measure the propagation channel, we conducted our measurement campaign in thirteenoutdoor and two indoor locations at the University of British Columbia (UBC), Vancouver campus.Fig. 3.4 shows all thirteen outdoor locations.The outdoor measurements were conducted in the Health Sciences area and adjacent parts ofthe UBC campus. This environment can be described as light urban. When we chose these outdoorlocations, we considered street-canyons, building facades, amount of vegetation, old and modernconstruction. Fig 3.5 (a), shows our measurement site 13, (indicated Fig. 3.4) and Fig. 3.5 (b),presents measurement site 2, where lots of vegetation were present around the transmitter andreceiver.The indoor data were collected in a hallway and a seminar room located on the fourth floor ofthe MacLeod building. Fig. 3.6 shows the measurement campaign in the hallway. All of thesemeasurements were conducted after office hours and weekends and restrictions were given on thepedestrians and vehicles to move through the channel. This allows us to model the channel usingrealistic data.In all cases, the channel was measured for D2D use cases. The position of the transmitter andreceiver were 1.15 m above the ground, which is a typical height for D2D applications. Channelresponse data were collected by placing the transmitter at a fixed position and by moving thereceiver along a continuous route, as shown in Fig. 3.5 (a) and (b), apart from the transmitter withdifferent step sizes such as 1 m, 2 m, 4 m, 5 m, for different measurement campaign.For each location of the receiver, we first measured the LOS channel by facing the receivertowards the transmitter. In order to measure our NLOS channel, we rotated the receiver in clock-wise, from -180 to +170 with 10 for outdoor and 3 for the indoor step to collect all MPCs which52Figure 3.4: Outdoor measurement sites.are coming from different angular spaces. Following this, we moved the receiver 1 m apart fromthe transmitter and repeated the same steps to collect all LOS and NLOS channel measurementdata. In case of NLOS channel measurement, received signals experience different propagationmechanisms such as reflection, diffraction, scattering which generated multipath components inthe propagation channel. These multipath signals can reach the receiver through single or multi-ple bounces from the reflectors. In the absence of LOS, it is essential to consider largest possibleMPCs. The minimum detectable signal of our channel sounder is -110 dBm. Thus, we hope tocapture all available MPCs.53(a)(b)Figure 3.5: Measurement sites, (a) without and (b) with vegetation.54Figure 3.6: Indoor measurement site (Hallway, wall to wall hallway width = 2.2m).3.3.3 Data CollectionWe measured channel frequency response in indoor and outdoor urban street canyon environmentswith our VNA based channel sounder. But for data analysis, we require the channel impulseresponse. For this purpose, we converted our frequency domain data to time domain throughproper post-processing. The details of the data acquisition/post-processing procedure are describedSection 2.2. We started data post-processing by passing measured data through a Hanning windowto reduce finite frequency span effect. We then applied inverse fast Fourier transform (IFFT) toconvert our frequency domain data into the time domain. The channel impulse response in thetime domain for jth location can be expressed as belowh j(t,f) =L jÂi=1a j,id (t t j,i)d (f f j,i), (3.9)55where summation over Lj represents the number of MPCs and a j,i, t j,i, f j,i, represent the complexpath gain, TOA and AOA of ith MPC in the jth receiver, respectively.3.3.4 Data ReductionAs described in the previous section, in the measurement campaign we measured complex wide-band channel frequency responses over full 360 spatial aperture. MPC parameters such as AOA,TOA and received power can be determined from the measured data by proper post-processing.Fig. 3.7(a) shows the distribution of received signal power observed by the first receiver; thecolour bar denotes the normalized received power. According to this figure, most of the receivedsignals are coming through LOS path. There are some significant NLOS paths as well, which re-sulted from the reflection and scattering of signals from different structures such as, building walls,metal frames of the building, poles, metal fence and so on. In the ideal case, each propagation pathis defined by a single ray. However, due to the surface roughness of the structures, reflected sig-nals are shown to be a combination of main reflection path and some sub paths which are closelyspaced to each other in both, time and angular domains [51]. A similar effect is also visible in ourmeasured data (Fig. 3.7 (a)). Closely spaced paths concentrated in the time domain is defined astime cluster, whereas, the angular spread of the path is defined as angular lobe [51].Fig. 3.7 (a) further indicates that each propagation path has some angular spread, which is dueto the distribution of electromagnetic field of the transmitter. The reason of this effect is illustratedin Fig. 3.8. From these two figures mentioned above, we observe that MPCs that are coming to theconsecutive receiving angles have almost similar TOA. We define these signals as an angular lobeor angular spread of a path. In MPC extraction we use one angular lobe to represent one MPC aslike the ideal case. Based on this principle, we perform beam cleaning.To estimate the MPC parameters, TOA, AOA and received power we used a CLEAN likealgorithm. The CLEAN algorithm was first introduced in [56]. This algorithm was previouslyused in radio astronomy, UWB [57], [58]. The CLEAN algorithm searches the received waveformiteratively to find the MPCs.56 Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Azimuth Angle of Arrival [deg]-150-100-500501001500.10.20.30.40.50.60.70.80.91(a)Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Azimuth Angle of Arrival [deg]-200-150-100-50050100150200(b)Figure 3.7: (a) Received power distribution and (b) extracted MPCs for 3 m transmitter-receiver separation as a function of TOA and AOA.57Transmitter ReceiverFigure 3.8: Effect of transmitting antenna beam pattern.The noise floor of our channel sounder is -110 dBm. We set a received power threshold of -100dBm to keep it 10 dB above the noise floor. We stored all data measured one angular position ina single vector. The side lobe of the receiver antenna appears in the time domain of the impulseresponse. We can remove this effect by using a variable threshold technique. From the radiationpattern of our horn antenna we observe that side lobe appears at 20 dB below the main beam, i.e.,the signal which is received by the side lobe is always 20 dB less than the main beam. Using thistechnique in the time domain, we can easily remove the side lobe effect of the receiving antennafrom the signal. The inherent disadvantage of this technique is that we will lose weak signals thatmay be coming from distant scatterers.After passing measurement data through the variable threshold and removing the sidelobe ef-fect of the antenna, the following algorithm is being used to extract MPCs:(i) Detect the index of strongest MPC from the channel impulse response, h(t).(ii) Save the associated TOA, AOA, and received power of detected MPC.(iii) Determine the point spread function of that MPC.58(iv) Subtract point spread function from channel impulse response h(t) and generate residualimpulse response hres(t).(v) Detect the index of strongest MPC from the residual impulse response hres(t).(vi) Find associated received power of that index and compare it with the received power thresh-old.(vii) If it is above the threshold then increase iteration step by one and go to step iii.Fig. 3.7 (b) shows all extracted MPCs using the above-mentioned algorithm, for transmitter-receiver separation of 3 m.3.4 Results3.4.1 Distance Metric SelectionFor persistence study, our first task is to determine whether persistent paths exist in mm-wavepropagation channel. To observe that, we plot the MPC for the whole receiver route in a singleplot. Fig. 3.9 shows all extracted MPCs for the entire receiver route where different markersrepresent the MPCs received by different receivers. However, this figure does not provide anyclear idea about which MPCs are persistent and which MPCs are not.This gives rise to the requirement of selection of a proper path tracking algorithm to connectMPCs logically. Before connecting MPCs, it is important to choose path tracking parameters prop-erly. In the previous works mentioned in Section 2.2, we have seen that parameters of persistentpaths depend on applications. For mm-wave applications, we choose TOA, AOA and receivedpower as path tracking parameter. However, before confirming these parameters as path track-ing parameter, we decide to see the performance of the distance metric, shown in Section 3.2, byusing different tracking parameters. To find the best distance metric to track all MPCs for the mm-wave system we tested three distance metrics: (i) one-dimensional (1D) distance metric, wherewe only use AOA as path tracking parameter (ii) two-dimensional (2D) distance metric, where we590 10 20 30 40 50 60 70 80 90 100Time of Arrival [ns]-200-150-100-50050100150200Azimuth Angle of Arrival [deg]3m separation4m separation5m separation6m separation7m separation8m separation9m separationFigure 3.9: Extracted MPCs of indoor scenario for receiver route 3-9 m (Hallway).use AOA and TOA as path tracking parameters, and (iii) three-dimensional (3D) distance metric,where we use AOA, TOA, and received power as path tracking parameters. In this work we do notinclude angle-of-departure (AOD) in our analysis. But this work can easily be extended to three-dimensional consideration which will include both AOD and AOA as path tracking parameters.3.4.1.1 1D distance metricFor path tracking using narrow band directional transceiver most important MPC parameter isAOA. So, first, we check 1D Euclidean Distance metric which only considers angular separation(AS) as a test of connectedness.As 1D metric only considers angular spread (AS), hence with moving receiver this metricconnect MPCs which have minimum AS. This is a subset of the distance metric presented atSection 3.2 and takes the form ofdk,i =q(fk( j1)fi( j))2. (3.10)60 Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Azimuth Angle of Arrival [deg]-200-150-100-500501001502003m separation4m separation5m separation6m separation7m separation8m separation9m separation(a)Transmitter-Receiver Separation [m]1 2 3 4 5 6 7 8 9 10Received Power [dBm]-100-90-80-70-60-50-40-30-20 Received power (Selected path)Regression line(b)Figure 3.10: (a) Persistent paths considering 1D distance metric (AOA only) and (b) receivedpower variation of a selected path (bold, red path) (Hallway).61Transmitter-Receiver Separation [m]1 2 3 4 5 6 7 8 9 10Received Power [dBm]-100-90-80-70-60-50-40-30-20 Received power (RT)Regression line(a)Transmitter-Receiver Separation [m]1 2 3 4 5 6 7 8 9 10Received Power [dBm]-100-90-80-70-60-50-40-30-20 Received power (RT)Regression line(b)Figure 3.11: Two simple examples of received power variation of persistent paths from raytracing simulated data.62However, when we analyze received power of the persistent paths that we find, some pathsshow a huge received power variation, one of such path is presented in Fig. 3.10(b). The boldblue line shows a -35 dB power variation for the receiver movement of 3 - 9 m, apart from thetransmitter. To check if this large power variation is normal, we generated two reflected pathsusing ray tracing simulation. In this simulation, we used two plain concrete reflectors and capturedreflected paths by moving the receiver apart from the transmitter with similar step size as measureddata. We generated the second reflected path by following the similar process, only changing theseparation between the transmitter-receiver route and the reflector wall.The reflected paths shown in Fig. 3.11 (a) and Fig. 3.11 (b). We compared the received powervariation of these two paths with the measured reflected path. Ray tracing simulated paths show8 dB and 12 dB variation in received power where the measured path shows a -35 dB variation,which reveals that the tested 1D distance metric is not the appropriate metric for persistent pathtracking. Further, to ensure the quality of the link, we need to track path in such a way that it showslower received power variation.3.4.1.2 2D distance metricTo avoid rapid power decay of a persistent path we incorporated power into the distance metric.This is a subset of the distance metric presented at Section 3.2 and takes the form ofdk,i =q(fk( j1)fi( j))2+(Prk( j1)Pri( j))2. (3.11)Here, another problem arises. This time, we are using AOA and received power in Euclideandistance metric. AOA and received power are in two different units and also their ranges, i.e.,the difference between the minimum and maximum, are different. If we use AOA and receivedpower in distance metric calculation without doing any normalization, the distance metric will bedominated by the AS than power separation. So, we decided to normalize measured data.Different normalization techniques are available for data normalization, such as (i) min-maxnormalization (ii) decimal scaling, and (iii) standard deviation method. However, the above men-63 Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Azimuth Angle of Arrival [deg]-200-150-100-500501001502003m separation4m separation5m separation6m separation7m separation8m separation9m separation(a)Transmitter-Receiver Separation [m]1 2 3 4 5 6 7 8 9 10Received Power [dBm]-100-90-80-70-60-50-40-30-20 Received power (1D distance metric)Received power (2D distance metric)Regression line (1D distance metric)Regression line (2D distance metric)(b)Figure 3.12: (a) Persistent paths considering 2D distance metric (AOA and received power)and (b) received power variation of the selected path (bold, blue path) (Hallway).64tioned normalization techniques are most suitable for normally distributed data. When the data isnot normally distributed, the Box-Cox transformation [59] is often used to transform non-normaldistributed data to normally distributed data. When we observe the distribution of AOA and re-ceived power of measured data, we find that AOAs are normally distributed where received powersare log-normally distributed. We connected the MPCs using 2D distance metric and results arepresented in Fig. 3.12 (a).The Received power is now included in the distance metric, so, MPC was connected based onminimum angular and power separation. Fig 3.12 (b) presents a comparison between 1D and 2Ddistance metric by comparing the same persistent paths connected by these two distance metrics.Results show that for the same receiver movement, the time received power variation of the pathwith distance travelled was only 10 dB, wherein the previous scenario this variation was 35 dB.We further compare the power variation of the measured reflected path with ray tracing simulatedreflected paths. We observe that during the lifetime of the path, power variation of the measuredpersistent path which was tracked using 2D distance metric is comparable with power variation ofray tracing simulated reflected paths.3.4.1.3 3D distance metricIn the previous work, where MPCs were tracked for V2V, and UWB applications, the authors com-pared MPC parameter, TOA, and AOA to connect persistent paths because these two parametersconnect MPCs which are generated from the same reflector. So, in the final distance metric, wedecided to use AOA, TOA and received power of MPCs in the distance metric as this will ensurethe persistent path is originating from the same scatterer/reflectors. The results are presented inFig. 3.13 (a).Fig. 3.13 (b) shows persistent paths which were connected using 1D and 3D metrics. Whenwe compare this result with Fig. 3.12 (b), we observe that including delay does not improvethe power fluctuation of a persistence path but incorporating delay in distance metric connectneighbour MPCs which are generated by a common reflector.65 Time of Arrival [ns]0 10 20 30 40 50 60 70 80 90 100Azimuth Angle of Arrival [deg]-200-150-100-500501001502003m separation4m separation5m separation6m separation7m separation8m separation9m separation(a).Transmitter-Receiver Separation [m]1 2 3 4 5 6 7 8 9 10Received Power [dBm]-100-90-80-70-60-50-40-30-20 Received power (1D distance metric)Received power (3D distance metric)Regression line (1D distance metric)Regression line (3D distance metric)(b)Figure 3.13: (a) Persistent paths considering 3D distance metric (AOA, TOA and receivedpower) and (b) received power variation of the selected path (Hallway).663.4.2 Persistence Paths Time of Arrival [ns]0 50 100 150 200 250 300 350Azimuth Angle of Arrival [deg]-200-150-100-500501001502004m separation6m separation8m separation10m separation12m separation14m separation16m separation18m separation20m separationMicro persistent MPCsMPC dropoutsFigure 3.14: Micro persistent MPCs and dropouts.We defined persistent paths in Section 3.2. In this subsection, we will see how persistent pathsevolve and change in the propagation environment with a moving receiver. Fig. 3.13 (a) showsall persistence paths, which are available on the measured propagation channel (Hallway). We seean LOS persistent path and some NLOS, reflected persistent paths with different lengths. Somepaths were persistent throughout the whole coverage area where some paths dropped briefly andappeared again. We also observe that the spatial-temporal variation of the path throughout itslength was not smooth, rather interrupted by the complexity of the wall.Fig. 3.14 presents persistent paths for another measurement site, site 11 (cf. Fig. 3.4). Thisfigure indicates two micro persistent paths. The first path (yellow) remains persistent for fourreceiver location, after that it disappears. The same path appears again (blue) and forms anotherpersistent path. According to the definition presented in Section 3.2, these two paths are the micropersistent path and the inactive period between these two paths is the dropout (indicated in the67figure). If we analyze the same path in large scale, these two individual paths will appear as asingle continuous path which is previously defined as macro persistent path because the large-scaleanalysis does not capture these temporary dropouts.3.4.3 Persistent Path StatisticsThis subsection presents the statistical characterizations of the persistence path length which werecaptured in the previous subsection. The variation of the persistent path can be characterizedstatistically by fitting the measurement data against theoretical distributions. Fig. 3.15 (a) presentsthe probability of occurrence of different lengths persistent paths which shows that about fourpercent of NLOS propagation paths were persistent throughout the whole coverage area. We alsohave some persistent paths with reduced life that disappeared after few meters due to the end oftheir corresponding propagation mechanism. Fig. 3.13 (a) also indicates that a huge percentage ofMPCs are appearing and disappearing with time, but not contributing to any persistent path.Due to the characteristics of the data, generally exponential distribution is used to model lowerfrequency persistent path lengths. However, because of the larger path decay of exponential dis-tribution, our data do not fit well in exponential distribution. In [45], log-logistic distribution wasused to model persistent path length. We find that this distribution also matches well with ourdata. Fig. 3.15 (a) presents, how well log-logistic PDF interprets the probability of occurrence ofdifferent lengths persistent paths. Log-logistic distribution has a heavier tail as like the empiricaldistribution of the data, which captures the persistent paths with a longer length. The spike inprobability for paths of length 7 m is due to the LOS path. Only considering NLOS paths wouldresult in an even better fit to the log-logistic distribution.To verify this distribution father, we tested goodness-of-fit using the Kolmogorov-Smirnov (KS)test [60] and find that measured data pass that test at the 0.05 significance level. After that, we com-pare cumulative distribution function (CDF) of empirical distribution and log-logistic distributionin Fig 3.15(b), where we see that although the empirical CDF is not a smooth curve because of alower number of samples, the trend fits well with the theoretical CDF.68Length of Persistent Path [m]0 1 2 3 4 5 6 7 8 9Probability of Occurance00.050.10.150.20.250.30.350.40.450.5Data Log-Logistic PDFµ = 0.573872 σ = 0.426635(a)Length of Persistent Path [m]0 1 2 3 4 5 6 7 8 9CDF00.10.20.30.40.50.60.70.80.91Empirical CDFLog-logistic CDF(b)Figure 3.15: (a) PDF and (b) CDF of persistent paths (Hallway).69The PDF of log-logistic distribution can be expressed asf (x;µ,s) = 1sxez(1+ ez)2;x> 0, (3.12)z=log(x)µs, (3.13)where µ is the scale parameter and s is the shape parameter. Before looking at how the parametersof log-logistic distributions are affected by the environment, first, we first consider the physicalsignificance of these parameters. Log-logistic scale parameter, µ , defines the log mean value ofthe distribution and can be used to calculate mean path length, lmean,lmean = log1(µ). (3.14)The log-logistic shape parameter, s , defines the dispersion of the distribution, i.e., the extentto which a distribution is stretched or squeezed. Dispersion of the distribution decreases as sdecreases.To examine if the log-logistic shape parameter, s , is related to propagation environment. Weplot the histogram of log-logistic shape parameter, s as shown in Fig. 3.16 (a). We find that forsome environments that shape parameters, s , are low, where for other locations its high, whichreveals that this parameter depends on propagation environment. For this analysis, we tested datafrom thirteen different outdoor urban locations and for the different arrangement of transmitter andthe receiver. We checked the relation of shape parameters with the separation of walls and amountof reflectors present in the environment.3.4.3.1 Wall separationFirst, we check the variation of shape factor for the different amount of wall separation, and wefind that wall separation has minimal effect on the parameters of the log-logistic distribution. This70result was expected because increasing/decreasing wall separation does not introduce any newMPC.3.4.3.2 Complexity indexWhen we check the variation of shape factor for the different amount of complexity, i.e., reflectorsand scatterers, present, we find that the variation of the shape factor is very small for differentenvironments and is between 0.3 and 0.47. Although the variation is small, with increasing amountof complexity in the propagation channel such as the complex structure of the reflecting surfaceand the presence of more reflector and scatterers, the channel introduces more persistent paths withdifferent lengths in the environment. This makes the distribution PDF more disperse, hence, theshape factor of the log-logistic distribution increases.To see this effect, we divide our thirteen different measurement locations into three categoriesand introduce an index to measure the amount of complexity in the environment. We define thisindex as complexity index. Three categories are (i) low, (ii) medium, and (iii) high complexityindex. From Fig. 3.16 (b) we see that light complexity in the propagation environment showslower shape factors where heavy complexity in the propagation environment provides higher shapefactors.The locations which fall under low complexity index, such as site 1, site 11, and site 7, onlyhave some distant reflectors and scattered trees. The location which falls under medium complexityindex, such as site 3, site 4, site 5 has two nearby walls with some street furniture such as poles,trees, etc. The location which falls under high complexity index such as site 2, site 6 and site13 have nearby reflectors in four directions, trees, and other street furniture. When we draw theregression line, we observe that low complexity index gives a shape factor of 0.33 where mediumand high complexity index provides shape factor of 0.39 and 0.44 respectively.3.4.3.3 Distance thresholdLog-logistic shape parameter, s also has a relation to the distance threshold, Thdist , which weuse as the threshold value to track persistent path. In order to observe how the distance threshold71Shape Factor, σ0.28 0.3 0.32 0.34 0.36 0.38 0.4 0.42 0.44 0.46 0.48Number of Occurance00.511.522.533.544.55(a)Complexity IndexLow Medium HighShape Factor, σ0.250.30.350.40.450.5Shape factorRegression line(b)Figure 3.16: (a) Histogram of the shape factor, s of log-logistic distribution (Hallway) and(b) relationship between the complexity of the propagation channel and log-logisticshape factor, s , double circles are representing overlapped points.720 5 10 15 20 25 30 35 40 45 50Distance Threshold, Thdist00.050.10.150.20.250.30.350.40.450.5Shape Factor, σShape factorRegression lineFigure 3.17: The relation between shape factor of log-logistic distribution and distancethreshold (Hallway).affects the parameters of the log-logistic distribution, we use different distance thresholds to formthe persistent path and observe the shape factors of corresponding log-logistic distributions. Theresult is presented in Fig 3.17, which shows that with increasing distance threshold, the shapefactor also increases. The reason is that increasing distance threshold generates more persistentpaths with different lengths, and so shape factor increases.From this figure, we also observe that the shape factor reaches a saturation point at distancethreshold of 30 and stayed similar when we increase the threshold to 35. At distance threshold 30all logical paths were connected. So, increasing threshold to 35 did not connect any new path. Butwhen we increase it to 40 some illogical connections are made. So, the valid distance threshold is30.73Fig. 3.18 (a) and (b) show histograms of mean path length for (a) 4 m step size (b) 2 m stepsize. From this figure, we see mean path length for 4 m step size is 5.7 m, where for 2 m step sizemean path length is 3.2 m.3.4.3.4 Relationship between received power and length of persistent pathsGenerally, a receiver chooses the best path based on the strength of the received signal. We ana-lyzed if there is any correlation between average received power and the length of the persistentpath. To see this correlation, we calculated the average received power of a persistent path andplotted according to their path length in Fig. 3.19. From the data trend, we see that higher sig-nal strength provides longer persistent paths because nearby continuous reflectors provide highersignal strength with longer path lengths. There are also some exceptions. From the figure, wesee some signal points, which have higher signal strength but low persistence. These signals aregenerating by nearby strong but discrete reflectors like metal frame, electric pole, lamp post, etc.However, most of the weak signals are arriving from the far scatterers which hold the relation.Some persistent paths have a weak signal for two possible reasons; (i) they are generating from farcontinuous scatterer and (ii) they are generating due to the presence of nearby discrete scatterers.We further calculate the correlation coefficient of the average received power and length ofthe persistent paths as the correlation coefficient of two random variables measures their lineardependence. We calculated the correlation coefficient of the length of a persistent path, l and theiraverage received power P of N observations, using Pearson correlation coefficient,r(l,P) = 1N1NÂi=1(liµlsl)(PiµPsP), (3.15)where µl and sl are the mean and standard deviation of persistent path length, l and µP and sP arethe mean and standard deviation of the average path power, P. The correlation coefficient for thissite is 0.51 which reveals that length of the persistent path exhibits weak, positive correlation withreceived signal strength.74Mean Path Length [m]4 4.5 5 5.5 6 6.5 7 7.5 8Number of Occurance00.511.522.533.544.55(a)Mean Path Length [m]2 2.5 3 3.5 4 4.5 5Number of Occurance00.511.522.533.54(b)Figure 3.18: Histograms of mean path length for, (a) 4m and (b) 2m step size.75Length of Persistent Path [m]0 1 2 3 4 5 6 7 8Average Received Power of Persistent Path [dBm]-75-70-65-60-55-50-45-40-35Average received power#Regression line r = 0.516318Figure 3.19: The relation between average received power and the length of the persistentpath at a specific scenario (Hallway).We further plot the histogram of the correlation coefficient for thirteen different locations inFig. 3.20 (a), which shows that for some locations, correlation coefficients are high where forother locations, correlation coefficients are low. This result shows that correlation coefficient isdependent on the propagation environment.3.4.3.5 Specular indexWe analyze environmental parameters that affect the correlation coefficient of average path powerand path length. We find that correlation coefficient is affected by specular index, i.e., amountof visibility of the reflecting wall/surface. With increasing amount of specular index correlationcoefficient also increases. Specular index of different environment can be different due to thepresence of vegetation which covers the main reflectors. Vegetation blocks the main reflectedpaths and reflects signal with lower signal strength. So, depending on the density of vegetation, itis possible to get a persistent path with the shorter length. In site 2 and site 7, shown in Fig. 3.4,76Correlation Coefficient, r0 0.1 0.2 0.3 0.4 0.5 0.6Number of Occurance00.511.522.533.544.55(a)Specular IndexLow Medium HighCorrelation Coefficient, r0.10.150.20.250.30.350.40.450.50.550.6Correlation coefficientRegression line(b)Figure 3.20: (a) Histogram of the correlation coefficient of the average received power andthe length of the persistent path and (b) their relation with specular index, for 12different measurement sites, double circles are representing overlapped points.77main reflectors were covered by the different amount of vegetation which provides low averagereceived power and length correlation coefficient.The specular index can also be reduced due to the type of reflectors, such as glass wall have alower specular index. For example, in site 11 and site 13, shown in Fig. 3.4, the main reflectorsare glass walls, thus in these sites, the average received power and length correlation coefficientare low. To see this effect, we divide thirteen different locations into three categories: (i) low, (ii)medium, and (iii) high specular index. From Fig. 3.20 (b) we see that low specular index in thepropagation environment shows lower correlation coefficient, where high specular index provideshigher correlation coefficient. When we draw a regression line, we found that low, medium andhigh specular index provide correlation coefficient of 0.25, 0.36 and 0.47 respectively.3.4.4 Angular Rate StatisticsTo track the spatial variation of MPCs during the receiver’s movement along its trajectory, weanalyze the rate of change of AOA, df/dx. The rate of arrival parameters can be easily estimatedfrom the measured data, as the accurate AOA are known for each MPC. To estimate AOA rate,the first arrival of the persistent path is considered to be the initial value. The arrival angle ofeach MPC within the path is subtracted from that of its immediate predecessor. The arrival rate ofAOA can be described by a random variable, and their probability density function fits well in anasymmetric Laplace distribution [61].Fig 3.21 (a) shows the probability of occurrence of different rate of AOA change, df/dx andtheoretical asymmetric Laplace PDF. Theoretical asymmetric Laplace pdf fits well the data. Inorder to find out the physical significance of why the rate of AOA change is Laplace distributed,we analyze the Fig. 3.13 (a) again and observe that reflected MPC, originated by a near reflector ischanging with a higher rate and the rate of AOA is even higher when the receiver and transmitterseparation is lower. Due to this fact, some higher rates of AOA change appeared in the histogram.On the other hand, most of the signals originate when transmitter and receiver separation is higher78dφ/dx-40 -30 -20 -10 0 10 20 30 40Probability of Occurance00.10.20.30.40.50.6 #µ = -1.87499 b = 7.26564 k = 1.19899RMS err = 0.0161DataLaplace PDF(a)dφ/dx-60 -40 -20 0 20 40 60CDF00.10.20.30.40.50.60.70.80.91Empirical CDFLaplace CDF(b)Figure 3.21: (a) PDF and (b ) CDF of Laplace distribution (Hallway).79and from the distance scatterer. This phenomenon provides a higher amount of rate of AOA occur-rence.We also calculate the root mean square (RMS) error of empirical and estimated Laplace PDFand find the error to be very small. For this specific case, the error is 0.0163. In addition to this,we compare the CDF of empirical data and Laplace distribution, as shown in Fig. 3.21 (b). TheCDF of the Laplace distribution matches well with empirical data.The asymmetric Laplace PDF can be written asf (x;µ,b,k) = 1b(k+1/k)e(xµ)sksb , (3.16)where s = sgn(x µ) , µ is a location parameter, b is a scale parameter and k is a asymmetryparameter. When k = 1, (xm)sks simplifies to |xm| and the distribution simplifies to thesymmetric Laplace distribution,f (x;µ,b) = 12be|xµ|b. (3.17)Laplace location parameter µ defines the mean value and asymmetry parameter k defines theasymmetry of the distribution. When we observe these two parameters for different propagationenvironments, we found these parameters to be less affected. The Laplace scale parameter b definesthe dispersion of the distribution, i.e., the extent to which a distribution is stretched or squeezed.Dispersion of the distribution decreases as b decreases. So the distribution is highly dependent onthis parameter and we studied this parameter in extensively.3.4.4.1 Laplace scale factorThe scale factor of the Laplace distribution is affected by the rate of AOA change, df/dx, ofthe path. In Section 3.2, it was observed that the rate of AOA change of persistent path, df/dx,is affected by the separation between receiver route and the nearest reflector. A similar effect is80Separation Between Main Reflector and Transmitter-Receiver Route [m]2 3 4 5 6 7 8 9Scale Factor, b11.522.533.544.5Scale factorRegression line(a)Distance Threshold, Thdist0 5 10 15 20 25 30 35 40 45 50Scale Factor, b012345678910Scale factorRegression line(b)Figure 3.22: (a) Laplace scale factor, b and the separation of the nearest wall from thetransmitter-receiver route and (b) Relationship between, Laplace scale factor, b anddistance threshold, Thdist . 81also observed in the measurement data. As this separation decreases, df/dx increases. With theincrease of df/dx, the scale factor, b, of Laplace distribution also increases.We plot the scale factor of Laplace distribution of five different locations and present the resultin Fig 3.22 (a). The relation between scale factor and wall separation shows a downward trendalthough points do not match perfectly with the regression line. The possible reason is that thepresence of different other reflectors and scatterers in the real scenarios.3.4.4.2 Distance threshold and Laplace scale factorThe distance threshold has a relation with the beamwidth of the antenna. With higher distancethreshold, two neighbouring MPCs with larger angular separation will be connected. These makeLaplace distribution more dispersed thus increases the shape factor of the distribution. Fig 3.22(b) shows that Laplace scale factor increases with the increasing distance threshold. For distancethreshold 25, it gets saturated. The similar distance threshold also saturated the log-logistic scalefactor. This result confirms that 25 is the most acceptable distance threshold for path trackingalgorithm.3.5 Application to SimulationThis section will propose a simulation model to incorporate finite persistence and angular variationinto ray tracing simulated data.3.5.1 Simulation of Persistence and Random ScattererWhen the receiver moves through a trajectory, the appearance and disappearance of MPCs causepersistence and dropouts. Although ray tracing simulation can be used to analyze the dynamic be-haviour of MPCs, it can only capture deterministic behaviour. In order to see the characteristics ofpersistent paths generated by ray tracing simulator, we use a very simplified parallel wall scenarioand generate LOS and two reflected paths, originated by two parallel walls, using Wireless InSite[62] software. Fig 3.23 (a) shows these three persistent paths, generated from ray tracing simu-lated data. These paths were generated deterministically so, all three paths persist for the whole820 50 100 150 200 250 300Time of Arrival [ns]-200-150-100-50050100150200Azimuth Angle of Arrival [deg]LOS pathRflected path 1Reflected path 2Simplified parallel wall scenario(a)0 50 100 150 200 250 300Time of Arrival [ns]-200-150-100-50050100150200Azimuth Angle of Arrival [deg]LOS pathReflected path 1Reflected path 2Added irregularity on the simplified wall(b)Figure 3.23: Persistent paths ray tracing simulation with (a) simplified parallel wall (b) acomplex wall on one side and simplified wall on the other side of the transmitter-receiver route.83 Time of Arrival [ns]0 50 100 150 200 250 300Azimuth Angle of Arrival [deg]-200-150-100-50050100150200LOS pathReflected path 1Reflected path 2(a) Time of Arrival [ns]0 50 100 150 200 250 300Azimuth Angle of Arrival [deg]-200-150-100-50050100150200Reflected path 1Reflected path 2LOS path(b)Figure 3.24: Persistent paths with (a) real AOA variation (b) real AOA variation and finitepersistence.84Figure 3.25: Semi-Markov model. Time of Arrival [ns]0 50 100 150 200 250 300Azimuth Angle of Arrival [deg]-200-150-100-50050100150200Figure 3.26: Persistent paths for a real parallel wall scenario including persistent path andrandom scatterers.85coverage area and show smooth spatio-temporal variation, i.e., all paths are macro persisted withfull-length.By introducing complexity in the geodata, we can generate realistic persistent paths. Usingsame Wireless InSite software, we introduced some irregularity, by adding several small beamswith a regular interval, on one wall and generate same LOS and two reflected path as before.Results are shown in Fig 3.23 (b). This time we see first reflected path is smooth as before, but inthe second reflected path, AOA change is not smooth, and rather than one micro persistent path,we see three micro persistent paths.3.5.2 Implementation of Realistic AOA VariationHowever, in the actual scenario, persistent paths do not evolve and change as in the case of raytracing simulated results. This is due to the complexity of the propagation environments. Fig.3.24 (a) shows persistent paths for a similar actual measurement site. Here, paths are not changingsmoothly in spatio-temporal domain. The complexity of the propagation introduces statisticalcomponents and generates more MPCs.Adding such details to the building outline data used by the ray tracing simulator is a time-consuming process. The alternative way is, use simplified building geometry to generate a per-sistent path, and then use a statistical model to introduce realistic AOA variation in the path. InSection 3.4, the rate of change of AOA of persistent paths follows a Laplace distribution. Thus,using this distribution, we can introduce realistic AOAs in the ray tracing generated persistentpaths.3.5.3 Implementation of PersistenceFig. 3.24 (b) shows the same persistent paths as shown in Fig. 3.24 (a) but this time we removedabrupt changes of paths by introducing finite persistence and dropouts. In order to incorporatesimilar persistence effect into conventional ray tracing results, we need a simulation model thatwill alternate between active and inactive states.86Previously in the literature, two finite state models were used to model birth and death of MPCs,(i) Markov model and (ii) semi-Markov model. Chong et al. [42] have used an M-step 4-stateMarkov channel model to model birth-death of MPCs for ultra-wideband applications. Markovmodel takes into account the time-varying properties of the channel by incorporating the dynamicevolution of paths when the receiver is in motion. However, there is a specific criterion whichneeds to meet before using Markov model. This model can only be used where state duration isexponentially distributed. In the above-mentioned application, the lifetime of MPCs was modelledusing the exponential distribution which meets the criteria of Markov model.In another work, which characterizes a time-varying MIMO channel, used a semi-Markovmodel to describe dynamic evolution of clusters [44]. The main advantage of a semi-Markovmodel is that this model is flexible regarding the distribution of state duration. In Fig. 3.15 (a) weobserve that the lifetime/persistence of MPCs is best explained by log-logistic distribution. So, wedecide to use the semi-Markov process to model dynamic evaluation of persistent paths.There are two-states in the semi-Markov model, active and inactive state, as shown in Fig. 3.25.The active state represents the continuation of a persistent path, while the inactive state representsa dropout. In Section 3.4, we confirmed that the probability of distribution of the active state wasa log-logistic distribution. The log-logistic distribution parameters can be chosen based on thedifferent complexity index of propagation environment from Fig. 3.16 (b). We do not estimate theprobability distribution for the inactive state due to lack of data with sufficiently fine evolution. Wepropose to use random numbers for dropouts. When the numerical value of dropout exceeds thehighest length of a persistent path, the path will not appear again.We could include the weak correlation of persistence length with the average received powerof MPCs. But given the very weak correlation and the considerable extra complexity, we chose notto do so. We do not believe that this will significantly impact the model fidelity.Finally, Fig. 3.26 shows persistent paths for a real parallel wall scenario. Fron this figure weobserve that along with real AOA variation and finite persistence there are some other persistentpaths and discrete MPCs which are coming from the random reflectors. We can easily introduce87these persistent paths and discrete MPCs using the METIS map-based model [2]. The proposedmodel can be implemented in the following manner.(i) Input ray tracing generated persistent path parameters, (ti,fi,Pri), length of ray tracing gen-erated path, Lrt and log-logistic and Laplace distribution parameters for a given propagationenvironment.(ii) Generate persistent path lengths, which are log-logistic random variables, for the given prop-agation environment using Eqn. 3.12.(iii) Generate AOA change of persistent paths, df , which are Laplace random variables, for thegiven propagation channel, using Eqn. 3.16.(iv) Choose a path length from the log-logistic random variables, assume the chosen path lengthis k, where k Lrt .(v) Change persistent path parameter, tk = ti, fk = fk1+dfi, Prk = Pri.(vi) If k 6= 0 then increase k and i by one and go to step v.(vii) If k = 0 then increase i by one and go to step vi.3.6 Impact of Persistence on System PerformancePersistence can affect throughput, beam training, and beam tracking procedure of the mm-wavesystem. In the following subsection, we will discuss the effect of persistence on different systemperformance.3.6.1 Throughput/CapacityStatistics of signal dropouts associate with finite persistence of MPC is similar to the burst errorstatistics. So, signal dropouts may affect system throughput/capacity as the similar way burst errordoes hence both can be modelled in a similar way. The Gilbert-Elliot model is used to model theburst error patterns in the transmission channel so, can also be used to model dropout problem.88The Gilbert-Elliot [65], [66] model is a two-state Markov model, where one state is defined as agood state which has low error probability and another state is defined as a bad state which has higherror probability. We propose to use Gilbert-Elliot model as it fits regarding physical significance.If we compare the states of Gilbert-Elliot model with persistence, we will see that when paths arepersistent, it will provide low error probability, and while dropouts occur it will show high errorprobability.3.6.2 Beam TrainingFirst mm-wave beam training scheme was standardized by IEEE 802.15.3c for wireless personalarea network (WPAN), where a two-stage beam training technique was used [63]. Some yearslater, another mm-wave beam training technique was standardized for wireless local area network(WLAN) by TG-ad)! (TG-AD)!)in the form of IEEE 802.11.ad, which was also based on two-stage beam training technique [64].Although no standards have been developed, in the recent years some works have been reportedin the literature regarding possible beam training schemes for 5G cellular networks [67–76]. Indynamic scenario device, all beam searching schemes are being developed based on the assumptionthat during beam training period device will remain static. In Fig. 3.14, we observed that inthe dynamic scenario the AOA of the beam changes moderately but due to the irregularity of thereflector/for deep fading sometimes temporary beam dropouts may take place. So, steps accountingfading or persistence need to include in beam training procedure.Here, we conjecture how finite persistence or occasional dropout will affect mm-wave beamtraining. The dropout may prevent a valid transmitter-receiver, beam combination from beingdetected. In two-stage beam training, in the first stage, transmitter and receiver search for the bestbeam using coarse beam and select best coarse beam pair. In the second stage, the coarse beamarea is being searched using fine beams. Persistence will affect both levels. The persistent problemcannot be detected if it occurs during the first stage of beam training. Because in the first stage, noinformation is available regarding the availability of a path. So, resolving persistence issue may be89more appropriate to level two where the strongest signal was detected but lost during the processdue to dropouts.Proper backoff time can be used to solve this problem. During this backoff time, fine beamtraining will be continued for the second best course beam, but it will check the strongest coursebeam pair to make sure if the dropped out paths appears again. To design proper backoff time werequire details data from proper field measurements where actual beam training will be interruptedby persistence. This necessitates developing a statistical model based on measurement data.However, the human presence or dynamic obstacles which block the path may have a similareffect. So, if the statistics of signal dropouts/persistence are characterized, it may be possible todwell or revisit possible transmitter-receiver beam combinations but this will certainly add consid-erable time in beam training process.3.6.3 Beam TrackingOnce beam training is done, a communication link between two links is established. In the dynamicscenarios, the AOA of path/beam changes due to: (i) movement of transmitter/receiver terminal,(ii) rotation of transmitter/receiver terminal, and (iii) change of environment. This necessitatestracking a link.Different approaches have been proposed to compensate for terminal rotation and translation.Some works have proposed to use sensors which are located inside the smart devices such as theaccelerometer and magnetometer to use in order to track beam [77, 78]. In [77], authors proposedto use the accelerometer, magnetometer and gyroscope sensors to predict device mobility andadapt/switch beams according to that. On the other hand, [78] proposed to use motion sensors suchas accelerometers, gyroscopes, and geomagnetic sensor as an attitude heading reference system(AHRS) and a zero-velocity detector (ZVD) to identify the cause of beam error and track beamaccording to that.Some other works have also been reported in the literature [79] and [80], where the authorsproposed to use a Kalman filter for beam tracking. Kalman filters are widely used in wireless90communication for the purpose of beam tracking. Zhang et al.[80] proposed to use a Kalman filterbased tracking algorithm to track the slow variation of AOA and AOD over time and also detectabrupt changes. Va et al. [80] proposed an extended Kalman filter (EKF) based algorithm, whichwill track beam by adapting with its AOA change over time. This types of algorithm will be ableto track the path until any drastic changes take place.Using beam tracking in the mm-wave system will reduce the frequency of time-consumingbeam training process. However, before successful implementation of such a beam tracking algo-rithm, we need to address several challenges. The first challenge is a measurement-based persis-tence model is required for the fair comparison of these beam tracking algorithms. Our proposedmodel in Section 3.5 will help to access the reliability of these tracking algorithms for differentscenarios.The second challenge is, in order to make the above-mentioned beam tracking algorithm fea-sible, the minimum beamwidth is required. Overhead of such beam tracking algorithm dependson the beamwidth of the antenna. Narrow beamwidth antenna will provide high beam trackingoverhead, while wide beamwidth antenna will have low antenna gain. So, there is a tradeoff be-tween these two factors. In Section 3.4 we have shown that the AOA change in faster rate whentransmitter and receiver separation is lower, so to track beam with low beam tracking overhead werequire wide beamwidth antenna. This is also acceptable regarding antenna gain although widebeam will prove lower antenna gain due to the smaller separation between transmitter and receiverthis antenna gain will be acceptable.On the other hand, AOA changes at a slower rate when transmitter and receiver separation ishigher, so to track a beam with low beamwidth antenna will provide acceptable beam trackingoverhead. Further, for a larger separation between transmitter and receiver higher antenna gain isrequired, which will be provided by this narrow beamwidth antenna. This discussion provides usintuition that adaptive beamforming antenna, which can also adapt the width of the beam, wouldbe the best option for the mm-wave system. More measurement and statistical modelling based onthis is required to confirm the appropriate range of adaptive beamformer beamwidth.91The third challenge is that beam tracking signal can sometimes change abruptly due todropouts, which will initiate a new beam training process. But, we have observed in Section 3,that when beam drops it often appears again after a brief interval. So, we propose to use properbackoff time and after that start beam tracking again. Again, more measurement is required todetermine the backoff time.3.7 DiscussionIn this chapter, we have shown that MPC persistence phenomenon that was previously observedat lower frequency also exist in the mm-wave frequency. To track persistent MPC, we used Eu-clidean distance metric. We found that 3D Euclidean distance metric is more appropriate to trackpersistence MPCs for mm-wave applications. Although most important parameters for mm-wavenarrow beamwidth systems beam tracking is AOA, including received power and TOA as pathtracking parameters, reduced the power variation of a persistent path and helped to connect MPCsoriginated by the similar scatterer, respectively.We have introduced the concept of macro and micro persistence and our results show that macropersistence could only explain large-scale signal variations, where micro persistence demonstratedsmall-scale signal variations, which are introduced due to the complexity of the real propagationenvironment, such as presence of window, beam, metal frame on the reflectors and vegetation, inthe form of dropouts and/or abrupt changes in terms of AOA.We have also characterized persistence statistics based on measurement data. Our observationsconfirmed that MPC persistence at mm-wave frequencies follows a log-logistic distribution aspreviously observed at lower frequencies and have yielded model parameters applicable to selectedoutdoor, indoor and hallway environments. We found that amount of complexity presented in theenvironment affect the shape factor of the log-logistic distribution by increasing the number ofdifferent lengths persistent paths. We also observed the correlation between path average powerand path length for thirteen locations and found a very weak correlation coefficient in the range of0.15-0.54. The correlation coefficient varied in this range due to its dependency on the specular92index of the environment. Higher specular index of the environment provides higher correlationcoefficient and vice verse. The rate of AOA change was also investigated and found that it followsa Laplace distribution. Further, it was found that how fast AOA will change greatly depends on therelative distance between a transmitter-receiver route and the nearby reflectors.In addition to this, we have proposed a persistence simulation model, based on our observationsthat can be used to make ray tracing simulations results more realistic for assessing the performanceof beam training and beam tracking schemes. Finally, we briefly explained how persistence wouldimpact system performance and some possible solutions which we hope will help system developerin developing successful mm-wave systems.93Chapter 4Summary and Conclusions4.1 ConclusionsOn the basis of the work reported in this thesis, we draw the following conclusions:Our 30 GHz VNA-based channel sounder is suitable for characterization of multipath persis-tence in indoor and outdoor environments. Our choice of the omnidirectional-directional antenna inthe channel sounding system was appropriate and accurately measured multipath components witha relatively lower measurement time than alternative schemes such as double-directional channelsounding.In both indoor and outdoor environments, we observed the existence of multipath persistenceat 30 GHz, which was previously observed for lower frequency channels. We further verifiedthat MPC persistence is well described by the log-logistic distribution that was previously usedto model persistence for the UWB (3-10 GHz) propagation channel. Our observations also showthat the log-logistic distribution parameter, s , depends on the amount of the complexity, such asthe presence of more complex walls, structures, building faces, and vegetation, in the propagationenvironment. Also, we found the existence of a weak correlation, between average persistent pathpower and path length and that this correlation depends on the amount of the visibility of thereflecting surface. Moreover, our results show that the angular variation of persistent paths are notdrastic and follows a Laplacian distribution with a good consistency. We also observed that the rate94of change of angular arrival (angular rate) of a persistent path substantially depends on the relativeseparation between reflectors and transmitter-receiver route. On this basis, we have proposed asemi-Markov model that can be used by designers to add realistic persistent paths to ray tracingsimulated data.4.2 Limitations and Future WorksThere are several limitations to this work. Firstly, we focused on D2D scenarios and only consid-ered the 2-dimensional channel; we did not consider D2I scenarios which would have made it nec-essary to consider elevation angle-of-arrival. This work can easily be extended to a 3-dimensionalchannel by including elevation angle of arrival in the distance metric which was explained in Sec-tion 3.2. Secondly, when we measured the propagation channel, the directional receiving antennawas rotated in steps of 3 for indoor measurement and 10 for outdoor measurement, not in a con-tinuous manner. Thirdly, during the measurement, we moved the receiver with a certain step size(1 m, 2 m, 4 m) rather than in a continuous manner. Finally, the receiver was always fixed. Onlythe transmitter was moved.The limitations of this study can be overcome in several ways. Finer step sizes in angle anddistance may be considered. Another possible direction is the development of a statistical model forMPCs dropouts, based on measurement data. This statistical model can be used to more accuratelydefine the inactive state of the semi-Markov persistent path simulation model. This model willfurther help to design the backoff time during the beam tracking phase of a mm-wave system.Another direction would be to carry out measurement campaigns in which both the transmitter andreceiver move, possibly along non-linear paths. The range of this rate will help to confirm therange of adaptive beamformer’s width of a mm-wave system, in its beam tracking phase.95Bibliography[1] “Technical feasibility of IMT in bands above 6 GHz,” ITU-R M.2376-0, Jul. 2015. ! pages1[2] “METIS channel models,” Mobile and wireless communications Enablers for the Twenty-twenty Information Society (METIS), Deliverable D1.4 ICT-317669-METIS/D1.4, Feb.2015. ! pages 1, 88[3] M. Peter et al., “Measurement campaigns and initial channel models for preferred suitablefrequency ranges,” Deliverable D2.1 mmMAGIC, Apr. 2016, [online] Available: https://5g-mmmagic.eu/results/deliverables. ! pages 1[4] “5G mmWave Channel Model Alliance,” [Online] Available: http://www.nist.gov/ctl/5g-mmwave-channel-model-alliance, 15 Sept. 2017. ! pages 1[5] IEEE Standard for Information technology– Local and metropolitan area networks– Specificrequirements– Part 15.3: Amendment 2: Millimeter-wave-based Alternative Physical LayerExtension,” in IEEE Std 802.15.3c-2009 (Amendment to IEEE Std 802.15.3-2003), pp.1 —200, Oct. 2009. ! pages 2[6] “SiBEAM,” [Online] Available: http://www.sibeam.com, 15 Sept. 2017. ! pages 2[7] “WirelessHD the first 60 GHz standard,” [Online] Available: http://www.wirelesshd.org, 15Sept. 2017. ! pages 2[8] IEEE Standard for Information technology–Telecommunications and information exchangebetween systems–Local and metropolitan area networks–Specific requirements-Part 11:Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) SpecificationsAmendment 3: Enhancements for Very High Throughput in the 60 GHz Band,” in IEEE Std802.11ad-2012 (Amendment to IEEE Std 802.11-2012, as amended by IEEE Std 802.11ae-2012 and IEEE Std 802.11aa-2012), pp.1 – 628, Dec. 2012. ! pages 3[9] “Federal Communications Commission,” FCC 16-89, Jul. 2016. ! pages 396[10] “3GPP,” [Online] Available: http://www.3gpp.org/release-15, 16 Sept. 2017. ! pages 3[11] A. F. Molisch et al., “Millimeter-wave channels in urban environments,” in Proc. EuCAP,2016, pp. 1 – 5. ! pages 4, 41[12] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wideband millimeter-wavepropagation measurements and channel models for future wireless communication systemdesign,” IEEE Trans. Commun., vol. 63, no. 9, pp. 3029 – 3056, Sept. 2015. ! pages[13] S. Hur et al., “Wideband spatial channel model in an urban cellular environments at 28 GHz,”in Proc. EuCAP, 2015, pp. 1 – 5. ! pages[14] X. Zhao, S. Li, Q.Wang, M.Wang, S. Sun, and W. Hong, “Channel measurements, model-ing, simulation and validation at 32 GHz in outdoor microcells for 5G radio systems,” IEEEAccess, vol. 5, pp. 1062 – 1072, 2017. ! pages[15] Y. Azar et al., “28 GHz propagation measurements for outdoor cellular communications usingsteerable beam antennas in NewYork City,” in Proc. ICC, Jun. 2013, pp. 5143-5147.! pages[16] R. Naderpour, J. Vehmas, S. Nguyen, J. Jrvelinen, and K. Haneda, “Spatio-temporal channelsounding in a street canyon at 15, 28, and 60 GHz,” in Proc. PIMRC, 2016, pp. 1 – 6. !pages[17] K. Haneda, S. L. H. Nguyen, J. Jrvelinen, and J. Putkonen, “Estimating the omni-directionalpathloss from directional channel sounding,” in Proc. EuCAP, 2016, pp. 1 – 5. ! pages[18] G. R. MacCartney, S. Deng, and T. S. Rappaport, “Indoor office plan environment and layout-based mmWave path loss models for 28 GHz and 73 GHz,” in Proc. VTC Spring, 2016, pp.1– 6. ! pages[19] S. Hur et al., “Proposal on millimeter-wave channel modeling for 5G cellular system,” IEEEJ. Sel. Topics Signal Process., vol. 10, no. 3, pp. 454 – 469, Apr. 2016. ! pages[20] N. Zhang, X. Yin, S. X. Lu, M. Du, and X. Cai, “Measurement-based angular characterizationfor 72 GHz propagation channels in indoor environments,” in Proc. GC Wkshps, 2014, pp.370 – 376. ! pages 4, 41[21] S. Sun et al., “Millimeter wave small-scale spatial statistics in an urban microcell scenario,”in Proc. ICC, May 2017, pp. 1– 7. ! pages 4, 4197[22] J. Wang, C. Gentile, P. B. Papazian, J. K. Choi, and J. Senic, “Quasi-deterministic modelfor Doppler spread in millimeter-wave communication systems,” IEEE Antennas WirelessPropag. Lett., vol. 16, pp. 2195 – 2198, 2017. ! pages 4, 8, 9, 11, 41[23] S. Salous, Radio Propagation Measurement and Channel Modelling, Hoboken, NJ,USA:Wiley, 2013. ! pages 7[24] Z. Wen and H. Kong, “MmWave MIMO channel sounding for 5G,” in Proc. 1st Int. Conf. 5GUbiquitous Connectivity, 2014, pp. 192 – 197. ! pages 8, 9[25] T. S. Rappaport et al., “Millimeter wave mobile communications for 5G cellular: It willwork!,” IEEE Access, vol. 1, pp. 335 – 349, 2013. ! pages 8, 9, 11[26] S. Hur, Y. J. Cho, J. Lee, Noh-Gyoung Kang, J. Park, and H. Benn, “Synchronous channelsounder using horn antenna and indoor measurements on 28 GHz,” in Proc. Int. Black SeaConf. Commun. Netw., May 2014, pp. 83 – 87. ! pages 11[27] R. J. Weiler et al., “Quasi-deterministic millimeter-wave channel models in MiWEBA,”EURASIP J. Wireless Commun. Netw., vol. 2016, no. 1, pp. 1 – 16, 2016. ! pages[28] K. Haneda, J. Jarvelainen, A. Karttunen, M. Kyro, and J. Putkonen, “A statistical spatio-temporal radio channel model for large indoor environments at 60 and 70 GHz,” IEEE Trans.Antennas Propag., vol. 63, no. 6, pp. 2694 – 2704, Jun. 2015. ! pages 8, 10, 11[29] C. U. Bas et al., “A real-time millimeter-wave phased array MIMO channel sounder,” [online]Available: https://arxiv:1703.05271v4. ! pages 8, 9, 10, 11[30] K. A. Remley et al., “Radio-wave propagation into large building structures - part 2: Charac-terization of multipath,” IEEE Trans. Antennas and Propag., vol. 58, no. 4, pp. 1290 – 1301,Apr. 2010. ! pages 10[31] G. Newhall, T. S. Rappaport, and D. G. Sweeney, “A spread spectrum sliding correlatorsystem for propagation measurements,” RF Design, Apr. 1996, pp. 40 – 54. ! pages 9[32] N.B. Carvalho, K.A. Remley, D. Schreurs, and K.G. Gard, “Multisine signals for wirelesssystem test and design,” IEEE Microwave Mag., June 2008, pp. 122 – 138. ! pages 9[33] G. R. MacCartney and T. S. Rappaport, “A flexible millimeter-wave channel sounder withabsolute timing,” IEEE J. Sel. Areas Commun., vol. PP, no. 99, pp. 1–13, 2017. ! pages 8,9, 1198[34] F. J. Harris, “On the use of windows for harmonic analysis with the discrete Fourier trans-form,” Proc. IEEE, vol. 66, no. 1, pp. 51 – 83, Jan. 1978. ! pages 25, 26[35] H. T. Friis, “A note on a simple transmission formula,” Proc. IRE, vol. 34, no. 5, pp. 254?256,May 1946. ! pages 31[36] J. E. Greivenkamp, Field Guide to Geometrical Optics, Bellingham, WA: SPIE Press, 2004,pp. 17–18. ! pages 34[37] A. Zajic et al., “Verification techniques for mmWave channel sounders,” 5G mmWave Chan-nel Model Alliance: Draft Rev. 1.2. ! pages 35[38] K. A. Remley, D. F. Williams, P. A. Hale, C-M Wang, J. Jargon, and Y. Park, “Millimeter-wave modulated-signal and error-vector-magnitude measurement with uncertainty,” IEEETrans. Microw. Theory Tech., vol. 63, no. 5, pp. 1710-1720, May 2015. ! pages 37[39] A. Newell, R. Baird, and P. Wacker, “Accurate measurement of Antenna Gain and Polariza-tion at Reduced Distances by an Extrapolation Technique,” IEEE Trans. Antennas Propag.,vol. 21, no. 4, July 1973. ! pages 37[40] T. Zwick, C. Fischer, D. Didascalou, and W. Wiesbeck, “A stochastic spatial channel modelbased on wave-propagation modeling,” IEEE J. Sel. Areas Commun., vol. 18, no. 1, pp. 6-15,Jan. 2000. ! pages 40[41] T. Zwick, C. Fischer, and W. Wiesbeck, “A stochastic multipath channel model includingpath directions for indoor environments,” IEEE J. Sel. Areas Commun., vol. 20, no. 6, pp.1178 – 1192, Aug. 2002. ! pages 40, 47[42] C. C. Chong, C. M. Tan, D. I. Laurenson, S. McLaughlin, M. A. Beach, and A. R. Nix,“A novel wideband dynamic directional indoor channel model based on a Markov process,”IEEE Trans. Commun., vol. 4, no. 4, pp. 1539 –1552, Jul. 2005. ! pages 40, 44, 47, 87[43] J. Nielsen, et al., “A dynamic model of the indoor channel,”Wireless Personal Commun., vol.19, no. 2, pp. 91–120, 2001. ! pages[44] Y. Chen and V. K. Dubey, “Dynamic simulation model of indoor wideband directional chan-nels,” IEEE Trans. Veh. Technol., vol. 55, no. 2, pp. 417 – 430, 2006. ! pages 87[45] F. O. Akgul and K. Pahlavan, “A model for spatial behavior of multipath components perti-nent to indoor geolocation using ray optics,” Int. J. Wireless Inf. Netw., vol. 20, no.4, pp. 328– 345, 2013. ! pages 40, 46, 47, 6899[46] K. Mahler, W. Keusgen, F. Tufvesson, T. Zemen, and G. Caire, “Tracking of wideband mul-tipath components in a vehicular communication scenario,” IEEE Trans. Veh. Technol., vol.66, no. 1, pp. 15-25, Jan. 2017. ! pages 40, 46[47] R. He et al., “A dynamic wideband directional channel model for vehicle-to-vehicle commu-nications,” IEEE Trans. Ind. Electron., vol. 62, no. 12, pp. 7870 – 7882, Dec. 2015. ! pages40[48] N. Czink, C. Mecklenbrauker, and G. Del Galdo, “A novel automatic cluster tracking algo-rithm,” in Proc. PIMRC, Sep. 2006, pp. 1– 5. ! pages 46, 50[49] S. Rangan, T. S. Rappaport and E. Erkip, “Millimeter-wave cellular wireless networks: po-tentials and challenges,” Proc. IEEE, vol. 102, no. 3, pp. 366-385, Mar. 2014. ! pages40[50] R. He et al., “Statistical characterization of dynamic multi-path components for vehicle-to-vehicle radio channels,” in Proc. VTC Spring, 2015, pp. 1– 6. ! pages 47[51] T. S. Rappaport, G. R. MacCartney, M. K. Samimi, and S. Sun, “Wideband millimeter-wavepropagation measurements and channel models for future wireless communication systemdesign,” IEEE Trans. Commun., vol. 63, no. 9, pp. 3029 – 3056, Sept. 2015. ! pages 56[52] A. Maltsev, A. Pudeyev, I. Karls, I. Bolotin, G. Morozov, R. Weiler, M. Peter, and W.Keusgen, “Quasi-deterministic approach to mmwave channel modeling in a non-stationaryenvironment,” in Proc. GC Wkshps, Dec. 2014, pp. 966 – 971. ! pages 40, 48[53] T. Santos, F. Tufvesson, and A. F. Molisch, “Modeling the ultra-wideband outdoor channel:Model specification and validation,” IEEE Trans. Commun., vol. 9, no. 6, pp. 1987–1997,Jun. 2010. ! pages 48[54] J. A. Hartigan and M. A. Wong, “Algorithm AS 136: a K-means clustering algorithm,” Ap-plied Statistics, vol. 28, no. 1, pp. 100 – 108, 1979. ! pages[55] T. Santos, J. Karedal, P. Almers, F. Tufvesson, and A. F. Molisch, “Modeling the ultra-wideband outdoor channel: Measurements and parameter extraction method,” IEEE Trans.Commun., vol. 9, no. 1, pp. 282-290, Jan. 2010. ! pages[56] J. A. Hogbom, “Aperture synthesis with a non-regular distribution of interferometer base-lines,” Astron. Astrophys. Suppl. Ser., vol. 15, 1974. ! pages 56100[57] R. J. M. Cramer, R. A. Scholtz, and M. Z. Win, “Evaluation of an ultra-wide-band propaga-tion channel,” IEEE Trans. Antennas Propag., vol. 50, no. 5, pp. 561– 570, May 2002. !pages 56[58] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, “Modeling the statisticaltime and angle of arrival characteristics of an indoor multipath channel,” IEEE J. Sel. AreasCommun., vol. 18, no. 3, pp. 347– 360, Feb. 2000. ! pages 56[59] G.E.P. Box and D.R. Cox, “An analysis of transformations,” J. Roy. Statist. Soc. B, vol. 26,pp. 211 – 252, 1964. ! pages 65[60] W. W. Daniel, Applied Nonparametric Statistics, 2nd ed., Boston: PWS-Kent., 1990, pp. 319– 330. ! pages 68[61] S. Kotz, T. J. Kozubowski, and K. Podgrski, The Laplace Distribution and Generalizations: ARevisit with Applications to Communications, Economics, Engineering and Finance., Basel:Birkhauser, 2001, pp. 23. ! pages 78[62] Wireless InSite., 2016, [online] Available: http://www.remcom.com/wireless-insite.! pages82[63] J. Wang, Z Lan et. al, “Beam codebook based beamforming protocol for multi-Gbpsmillimeter-wave WPAN systems,” IEEE J. Sel. Areas Commun., vol. 27, no. 8, pp. 1390–1399, Oct. 2009. ! pages 89[64] E. Perahia, C. Cordeiro, M. Park, and L. L. Yang, “IEEE 802.11ad: Defining the next gen-eration multi-Gbps Wi-Fi,” in Proc. 7th IEEE Consum. Commun. Netw. Conf., pp. 1– 5, Jan.2010. ! pages 89[65] E. N. Gilbert, “Capacity of a burst-noise channel,” Bell Syst. Tech. J., vol. 39, no. 5, pp. 1253–1265, Sept. 1960. ! pages 89[66] E.O. Elliott, “Estimates of error rates for codes on burst-noise channels,” Bell Syst. Tech. J.,vol. 42, pp. 1977 –1997, 1963. ! pages 89[67] M. Jasim and N. Ghani, “Fast beam discovery for mmWave cellular networks,” in Proc.WAMICON, 2017, pp. 1– 6. ! pages 89[68] A. Alkhateeb, Y. H. Nam, M. S. Rahman, J. Zhang, and R.W. Heath, “Initial beam associationin millimeter wave cellular systems: Analysis and design insights,” IEEE Trans. WirelessCommun., vol. 16, no. 5, pp. 2807 – 2821, May 2017. ! pages101[69] J. Kim and S.-N. Hong, “Dynamic two-stage beam training for energy efficient millimeter-wave 5G cellular systems,” Telecommun. Syst., vol. 59, no. 1, pp. 111 – 122, 2014. ! pages[70] M. E. Eltayeb, A. Alkhateeb, R. W. Heath, and T. Y. Al-Naffouri, “Opportunistic beam train-ing with hybrid analog/digital codebooks for mmWave systems,” in Proc. GlobalSIP, Dec.2015, pp. 315-319. ! pages[71] A. Alkhateeb, O. El Ayach, G. Leus, and R. W. Heath, “Hybrid precoding for millimeterwave cellular systems with partial channel knowledge,” in Proc. ITA, Feb. 2013, pp. 1-5. !pages[72] M. Giordani, M. Mezzavilla, C. N. Barati, S. Rangan, and M. Zorzi, “Comparative analysisof initial access techniques in 5G mmWave cellular networks,” in Proc. CISS, 2016, pp. 268– 273. ! pages[73] M. Giordani, M. Mezzavilla, and M. Zorzi, “Initial access in 5G mmWave cellular networks,”IEEE Commun. Mag., vol. 54, no. 11, pp. 40 – 47, Nov. 2016. ! pages[74] C. N. Barati et al., “Initial access in millimeter wave cellular systems,” IEEE Trans. WirelessCommun., vol. 15, no. 12, pp. 7926 – 7940, Dec. 2016. ! pages[75] A. Capone, I. Filippini, and V. Sciancalepore, “Context information for fast cell discovery inmm-Wave 5G networks,” in Proc. European Wireless Conf., May 2015, pp. 1 – 6. ! pages[76] W. B. Abbas and M. Zorzi, “Context information based initial cell search for millimeter wave5G cellular networks,” in Proc. EuCNC, 2016, pp. 1 – 7. ! pages 89[77] Z. Yang, P. H. Pathak, Y. Zeng, and P. Mohapatra, “Sensor-assisted codebook-based beam-forming for mobility management in 60 GHz WLANs,” in IEEE 12th Int. Conf. Mobile AdHoc Sensor Syst., Dallas, 2015, pp. 333 – 341. ! pages 90[78] D.S. Shim, C.K. Yang, J. H. Kim, J. P. Han, and Y. S. Cho, “Application of motion sensorsfor beam-tracking of mobile stations in mmWave communication systems,” Sensors, vol. 14,no. 10, pp. 19622-19638, Oct. 2014. ! pages 90[79] C. Zhang, D. Guo, and P. Fan, “Tracking angles of departure and arrival in a mobile millimeterwave channel,” in Proc. ICC, May 2016, pp. 1– 6. ! pages 90[80] V. Va, H. Vikalo, and R. W. Heath, “Beam tracking for mobile millimeter wave communica-tion systems,” in Proc. GlobalSIP, 2016, pp. 743 – 747. ! pages 90, 91102[81] Y. Inoue, Y. Kishiyama, Y. Okumura, J. Kepler, and M. Cudak, “Experimental evaluation ofdownlink transmission and beam tracking performance for 5G mmW radio access in indoorshielded environment,” in Proc. PIMRC, Aug. 2015, pp. 862 – 866. ! pages103