You may notice some images loading slow across the Open Collections website. Thank you for your patience as we rebuild the cache to make images load faster.

Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Configuration of multiple input multiple output antenna arrays for wireless communications in underground… Emami Forooshani, Arghavan 2013

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

Notice for Google Chrome users:
If you are having trouble viewing or searching the PDF with Google Chrome, please download it here instead.

Item Metadata


24-ubc_2013_fall_emamiforooshani_arghavan.pdf [ 2.22MB ]
JSON: 24-1.0074200.json
JSON-LD: 24-1.0074200-ld.json
RDF/XML (Pretty): 24-1.0074200-rdf.xml
RDF/JSON: 24-1.0074200-rdf.json
Turtle: 24-1.0074200-turtle.txt
N-Triples: 24-1.0074200-rdf-ntriples.txt
Original Record: 24-1.0074200-source.json
Full Text

Full Text

CONFIGURATION OF MULTIPLE INPUT MULTIPLEOUTPUT ANTENNA ARRAYS FOR WIRELESSCOMMUNICATIONS IN UNDERGROUND MINESbyArghavan Emami ForooshaniB.Sc., Iran University of Science and Technology, 2000M.Sc., University of Manitoba, 2006A THESIS SUBMITTED IN PARTIAL FULLFILLMENT OFTHE REQUIREMENTS FOR THE DEGREE OFDOCTOR OF PHILOSOPHYinThe Faculty of Graduate and Postdoctral Studies(Electrical and Computer Engineering)THE UNIVERSITY OF BRITISH COLUMBIA(Vancouver)August 2013? Arghavan Emami Forooshani, 2013iiABSTRACTIn recent years, the underground mining community has begun to embrace standards-basedshort-range wireless communications technology as a key part of their strategy for enhancingthe safety and productivity of their operations. Here, we show how the significant differencesbetween wireless propagation in conventional surface environments and underground minesaffect the design of modern wireless communications systems based upon multiple-inputmultiple-output (MIMO) antenna array technology. In order to achieve this goal, we haveemployed a variety of approaches to characterize wireless propagation (and MIMO-basedwireless system performance) in underground environments representative of those found inmodern hard rock mines, including: 1) field measurements collected using a custom-designedchannel sounder in both a building service tunnel at the University of British Columbia andan underground lead-zinc mine at Myra Falls, BC, 2) simulations based upon ray-tracing inrepresentative environments and 3) theoretical models based upon waveguide modeexpansion in representative environments. We have used the results obtained: 1) to determinethe reduction in the angular spread of multipath signals that arrive at the receiver in anunderground mine compared to that observed in conventional surface environments and themanner in which it decreases with increasing transmitter-receiver separation and 2) to showthat the antenna elements in MIMO antenna arrays used in underground environments musttherefore be separated by several wavelengths (rather than the customary half-wavelengthused in surface environments) in order to achieve acceptable performance. Further, theseparation between the antennas must increase as the transmitter-receiver separationincreases, higher order modes attenuate and, as a consequence, angular spread decreases.Other outcomes of this work include: 1) demonstration that the power azimuth spectrum(PAS) in underground mine environments can be modeled by a Gaussian distribution and2) development of a novel technique based upon particle swarm optimization (PSO) forassessing and optimizing the performance of distributed-MIMO antenna systems inunderground mine environments.iiiPREFACEThis thesis presents research conducted by Arghavan Emami-Forooshani under thesupervision of Prof. David G. Michelson in the Radio Science Lab (RSL) at the University ofBritish Columbia, Vancouver campus. Prof. Sima Noghanian (University of North Dakota)provided valuable technical suggestions and feedback during various stages of the thesisproject.A version of Chapter 2, ?A survey on wireless propagation modeling in undergroundmines? has been accepted for publication in IEEE Communications Surveys and Tutorials[manuscript ID: COMST-00130-2012-R1]. Prof. Sima Noghanian read an early draft andprovided valuable technical feedback. Shahzad Bashir suggested some revisions to theIntroduction and Conclusions.During the development of the MIMO channel sounder in Chapter 3, Robert D. Whiteassisted with development of the data acquisition code.A poster based on Chapter 4, ?Effect of antenna configuration on performance ofMIMO-based access points in a service tunnel? was presented at IEEE APS/USNC-URSI2012(Chicago, IL), in Jul. 2012 and an abstract was published in the conference proceedings.A version of Chapter 5, ?Effect of antenna array properties on MIMO systemperformance in an underground mine,? has been accepted for publication in IET Microwave,Antennas and Propagation [manuscript ID: MAP-2013-0102.R1]. Robert D. Whiteparticipated in the MIMO measurement campaign in the underground mine.ivFor Chapter 6, Prof. Sima Noghanian provided valuable technical feedbackthroughout the project.vTABLE OF CONTENTSAbstract..................................................................................................................................... iiPreface..................................................................................................................................... iiiTable of Contents .................................................................................................................... iiiList of Tables ......................................................................................................................... viiiList of Figures ......................................................................................................................... ixAbbreviations.......................................................................................................................... xiiAcknowledgements..................................................................................................................xvDedication............................................................................................................................. xviiCHAPTER 1: INTRODUCTION AND OVERVIEW ............................................................ 11.1 Wireless in Underground Mines........................................................................................ 21.2 Characterization of MIMO-Based Systems...................................................................... 51.3 Previous Efforts to Design Antenna Configuration in Indoor Environments ............... 81.4 Implications for MIMO Antenna Configuration Design and Deployment inUnderground Mines....................................................................................................................... 111.5 Objectives of This Work................................................................................................... 121.6 Organization of the Thesis ............................................................................................... 12CHAPTER 2: A SURVEY OF WIRELESS COMMUNICATIONS AND PROPAGATIONMODELING IN UNDERGROUND MINES ....................................................................... 142.1 Introduction....................................................................................................................... 142.2 Wireless Propagation Terminology................................................................................. 172.3 The Evolution of Wireless Communications in Tunnels and Underground Mines .... 222.4 Propagation Analysis and Modeling ............................................................................... 29vi2.5 Measurement-Based Modeling ........................................................................................ 492.6 Implications for Wireless Communication System Design ........................................... 602.7 Conclusions........................................................................................................................ 71CHAPTER 3: MIMO EXPERIMENTAL SETUP AND DATA COLLECTION ............... 743.1 Introduction....................................................................................................................... 743.2 Development of a MIMO Channel Sounder................................................................... 74CHAPTER 4: EFFECT OF ANTENNA CONFIGURATION ON MIMO-BASEDACCESS POINTS IN A SHORT TUNNEL ......................................................................... 814.1 Introduction....................................................................................................................... 814.2 Simulation Setup and Scenarios ...................................................................................... 844.3 Multiple Antenna Analysis............................................................................................... 894.4 Simulation Results and Discussions................................................................................. 914.5 Experimental Validation ................................................................................................ 1024.6 Conclusions...................................................................................................................... 105CHAPTER 5: EFFECT OF ARRAY PROPERTIES ON MIMO SYSTEMPERFORMANCE IN AN UNDERGROUND MINE ........................................................ 1075.1 Introduction..................................................................................................................... 1075.2 MIMO Measurement in an Underground Mine .......................................................... 1105.3 Multiple Antenna Analysis............................................................................................. 1155.4 Measurement Results and Discussions.......................................................................... 1185.5 Conclusions...................................................................................................................... 129CHAPTER 6: CHARACTERIZATION OF ANGULAR SPREAD IN UNDERGROUNDTUNNELS BASED ON A MULTIMODE WAVEGUIDE MODEL................................. 1316.1 Introduction..................................................................................................................... 1316.2 A Multimode Waveguide Model.................................................................................... 133vii6.3 Angular Spread and Capacity Characterization Based on a Multimode WaveguideModel ........................................................................................................................................... 1356.4 Results and Validation of Multimode Modeling .......................................................... 1386.5 Extension of the IEEE 802.11n MIMO Channel Model to Underground Mines ...... 1536.6 Conclusions...................................................................................................................... 154CHAPTER 7: OPTIMIZATION OF ANTENNA PLACEMENT IN DISTRIBUTEDMIMO SYSTEMS FOR UNDERGROUND MINES......................................................... 1567.1 Introduction..................................................................................................................... 1567.2 Multimode Waveguide Modeling and Experimental Validation................................ 1597.3 Optimization of C-MIMO and D-MIMO Configurations........................................... 1627.4 Comparison of C-MIMO and D-MIMO System Performance................................... 1687.5 Conclusions...................................................................................................................... 178CHAPTER 8: CONCLUSIONS .......................................................................................... 1818.1 Contributions................................................................................................................... 1818.2 Recommendations for Future Work ............................................................................. 183REFERENCES .................................................................................................................... 185viiiLIST OF TABLESTABLE 2-1 OVERALL LOSS ALONG A STRAIGHT PATH (EH MODE WITH HALF-WAVE ANTENNAS) [11]. .................38TABLE 2-2 OVERALL LOSS ALONG A PATH INCLUDING ONE CORNER (EH MODE WITH HALF-WAVE ANTENNAS)[11]. ...........................................................................................................................................................38TABLE 2-3 COMPARISON OF DIFFERENT ANALYTICAL MODELS............................................................................48TABLE 2-4 SOME SIMILARITIES AND DIFFERENCES BETWEEN UNDERGROUND MINES, STRAIGHT LONG TUNNELSAND CONVENTIONAL INDOOR FOR PROPAGATION AT UHF-BAND (F: FREQUENCY, N: PATHLOSS EXPONENT,?RMS: RMS DELAY SPREAD, DTX-RX: TX AND RX DISTANCE, ?: INCREASE, ?: DECREASE) ................................52TABLE 2-5 PATHLOSS EXPONENT AND DELAY SPREAD ASSUMING OMNI-DIRECTIONAL ANTENNAS FOR SEVERALFREQUENCIES IN DIFFERENT ENVIRONMENTS. ..............................................................................................55TABLE 2-6 KEY IMPLICATIONS FOR WIRELESS SYSTEM DESIGN AT UHF-BAND IN UNDERGROUND MINES. ..........71TABLE 3-1 MEASUREMENT SPECIFICATIONS ........................................................................................................76TABLE 4-1 SPECIFICATIONS OF THE MAIN STRUCTURES USED IN THE WIRELESS INSITE SIMULATIONS. ...............86TABLE 4-2 DIFFERENT ANTENNA CONFIGURATION AND DEPLOYMENT SCENARIOS (X: PARALLEL TO THE TUNNELWIDTH, Y: PARALLEL TO THE TUNNEL HEIGHT, LONG: PARALLEL TO THE TUNNEL AXIS, H: HORIZONTALPOLARIZATION AND V: VERTICAL POLARIZATION). .....................................................................................88TABLE 4-3 CORRELATION COEFFICIENT OF NEIGHBORING ELEMENTS AT RX WITH DIFFERENT INTERELEMENTSEPARATIONS (V: VERTICAL, H: HORIZONTAL POLARIZATIONS). .................................................................93TABLE 4-4 AVERAGE PATHLOSS AT 2.49 GHZ. ..................................................................................................100TABLE 4-5 MEASUREMENT SPECIFICATIONS ......................................................................................................102TABLE 4-6 CAPACITY STATISTICS AT 2.49 GHZ AND SNR=20 DB TO COMPARE MIMO CAPACITY OF DIFFERENTANTENNA CONFIGURATIONS (WITH POWER-IMPACT) .................................................................................103TABLE 5-1 MIMO MEASUREMENT SCENARIOS IN THE MYRA FALLS UNDERGROUND MINE. ..............................114TABLE 5-2 ENVELOPE CORRELATION COEFFICIENTS OF RX GRID ANTENNAS FOR DIFFERENT MEASUREMENTSCENARIOS (F=2.49 GHZ)..........................................................................................................................115TABLE 5-3 SINGULAR VALUES OF MEASURED H-MATRICES AND I.I.D. RAYLEIGH H-MATRICES (ALL SINGULARVALUES ARE NORMALIZED TO THE LARGEST ONE). ....................................................................................121TABLE 5-4 MEAN PATHLOSS MEASURED AT EACH RX GRID LOCATION IN MYRA-FALLS MINE (AT 2.49 GHZ). .128TABLE 6-1 ANGULAR SPREADS (DEG) FOR DIFFERENT ZONES OF THE LARGE AND SMALL TUNNEL OBTAINED BYMULTIMODE AND RAY-TRACING METHODS. ...............................................................................................142TABLE 6-2 COMPARISON OF AS FOR DIFFERENT ENVIRONMENTS. ....................................................................146TABLE 6-3 MEDIAN AND 10% OUTAGE 4?4 MIMO CAPACITY (BIT/SEC/HZ) AT SNR=20 DB FOR SEVERALINTERELEMENT SPACINGS AND DIFFERENT ZONES OF THE SMALL TUNNEL.................................................153TABLE 7-1 MULTIMODE AND PSO SIMULATION SETUP ......................................................................................169ixLIST OF FIGURESFIGURE 1-1 TECHNOLOGY PUSH INTO MINING UNDERGROUND MINES OVER TIME. .............................................4FIGURE 1-2 DIFFERENT TYPES OF MIMO ANTENNA CONFIGURATIONS, SUCH AS ULA AND UCA FOR INDOORENVIRONMENTS. ..........................................................................................................................................10FIGURE 1-3 FIXED TX ULA PLACED ON THE PLATFORM, MOBILE RX ULA INSTALLED ON THE TRAINWINDSHIELD [21]. ........................................................................................................................................11FIGURE 2-1 WIRELESS PROPAGATION PHENOMENA..........................................................................................18FIGURE 2-2 LARGE-SCALE AND SMALL-SCALE FADINGS ..................................................................................19FIGURE 2-3 A PATHLOSS MODEL WITH THREE BREAKPOINTS ...........................................................................21FIGURE 2-4 THROUGH-THE-EARTH COMMUNICATIONS ....................................................................................24FIGURE 2-5 LEAKY FEEDER CABLE...................................................................................................................26FIGURE 2-6 TRACKING SYSTEM IN AN UNDERGROUND MINE [52].....................................................................27FIGURE 2-7 SURFACE ANTENNA FOR THROUGH-THE-EARTH COMMUNICATIONS. .............................................30FIGURE 2-8 A) CURRENT DISTRIBUTION AND ELECTRIC FIELD LINES OF THE MONOFILAR MODE (BETWEEN EACHWIRE AND THE TUNNEL WALL) AND B) MONOFILAR MODE AND BIFILAR MODE (BETWEEN THE WIRES) [30].32FIGURE 2-9 COMPARISON OF REFLECTION ANGLES OF LOWER AND HIGHER ORDER MODES IN A WAVEGUIDE. .34FIGURE 2-10 MAP AND DIGITAL PHOTOGRAPH OF AN UNDERGROUND GALLERY [60]. .......................................35FIGURE 2-11 OVERALL LOSS FOR VARIOUS DISTANCES ALONG A STRAIGHT TUNNEL FOR HALF-WAVE ANTENNASWITH HORIZONTAL POLARIZATIONS (Z IS TX-RX DISTANCE) [11]. ...............................................................37FIGURE 2-12 MEASURED DATA FOR TWO POLARIZATIONS AT 900 MHZ TOGETHER WITH THE TWO-SLOPEREGRESSION FITS IN A COAL MINE [74]. .......................................................................................................41FIGURE 2-13 STOCHASTIC SCATTERING APPROACH: REFLECTION AT RANDOMLY ORIENTED TANGENTIAL PLANESFOR EACH DISCRETE RAY (A) SAME PLANE FOR ALL THE RAYS AND (B) RANDOMLY ORIENTED PLANES [87]. ........................................................................................................................................................47FIGURE 2-14 (A) A RECTANGULAR WAVEGUIDE AND (B) CIM MODEL FOR THE WAVEGUIDE IN FIGURE (A) [88]. ........................................................................................................................................................47FIGURE 2-15 RADIUS OF A CURVATURE IN A RECTANGULAR CURVED TUNNEL [13]. ..........................................64FIGURE 3-1 BLOCK DIAGRAM OF THE MEASUREMENT SETUP............................................................................76FIGURE 3-2 MIMO MEASUREMENT EQUIPMENT IN RSL..................................................................................77FIGURE 3-3 CONNECTION DIAGRAM FOR CALIBRATION....................................................................................79FIGURE 3-4 FLOWCHART OF THE UWB-MIMO CHANNEL SOUNDER SOFTWARE..............................................80FIGURE 4-1 UBC WOODWARD SERVICE TUNNEL CONSTRUCTED IN WIRELESS INSITE WITH ITS EXTENSIVEINFRASTRUCTURE AND CONSIDERING 3 PROPAGATION SCENARIOS. .............................................................85xFIGURE 4-2 ANTENNA CONFIGURATION SCENARIOS USED IN THIS STUDY (V: VERTICAL, H: HORIZONTAL). ....89FIGURE 4-3 COMPARISON OF DIFFERENT ANTENNA CONFIGURATIONS FOR THE TUNNEL WITHOUTINFRASTRUCTURE. .......................................................................................................................................96FIGURE 4-4 COMPARISON OF THE PERFORMANCE OF TR-X-V AND TR-Y-H ANTENNA CONFIGURATIONS AT RX2.......................................................................................................................................................96FIGURE 4-5 COMPARISON OF DIFFERENT ANTENNA CONFIGURATIONS FOR THE TUNNEL WITHINFRASTRUCTURE. .......................................................................................................................................97FIGURE 4-6 IMPACT OF INFRASTRUCTURE ON PERFORMANCE OF TR-X-V-SC6 ANTENNA CONFIGURATION. .....99FIGURE 4-7 COMPARISON OF SIMULATED (WITH INFRASTRUCTURE) AND MEASURED MIMO CAPACITY CDFS. ......................................................................................................................................................104FIGURE 5-1 MAP OF THE MYRA-FALLS MINE IN B.C., CANADA AND THE TRANSMITTING (TX) ARRAY ANDRECEIVER (RX) GRID LOCATIONS (5 TX ARRAY LOCATIONS AND 2 RX GRID LOCATIONS). .........................111FIGURE 5-2 PHOTOGRAPHY OF RF EQUIPMENT IN MYRA FALLS MINE. ..........................................................111FIGURE 5-3 SPATIAL CORRELATION ANALYSIS ON THE SUCCESSIVE ANTENNA ELEMENTS ON THE RX GRID FORTWO DIFFERENT ARRAY ORIENTATIONS: (A) PERPENDICULAR TO THE TUNNEL AXIS AND (B) PARALLEL TOTHE TUNNEL AXIS. .....................................................................................................................................114FIGURE 5-4 CDFS OF 4?4-MIMO CAPACITY WITHOUT POWER CONSIDERATIONS (BASED ON THEMEASUREMENT).........................................................................................................................................119FIGURE 5-5 ANGULAR SPREAD VARIATION VERSUS DISTANCE (BASED ON MULTIMODE WAVEGUIDE MODEL). ........................................................................................................................................................122FIGURE 5-6 CAPACITY OF 4?4-MIMO SYSTEM FOR DIFFERENT ANTENNA SPACINGS (BASED ON THEMULTIMODE WAVEGUIDE MODEL). ............................................................................................................123FIGURE 5-7 CDFS OF 4?4-MIMO CAPACITY WITH POWER CONSIDERATIONS (BASED ON THE MEASUREMENT). ......................................................................................................................................................124FIGURE 5-8 A TUNNEL WITH RECTANGULAR CROSS-SECTION.........................................................................126FIGURE 5-9 CDFS OF 4?4-MIMO CAPACITY WITH POWER CONSIDERATIONS (BASED ON MEASUREMENT ANDMULTIMODE WAVEGUIDE MODEL AT F=2.49 GHZ, SNR=20 DB). .............................................................129FIGURE 6-1 COMPARISON OF MULTIMODE MODEL WITH EXPERIMENTAL WORK OBTAINED FROM [13]...........135FIGURE 6-2 REFLECTION ANGLES OF LOWER AND HIGHER ORDER MODES FROM SIDEWALLS IN A WAVEGUIDE(?1 IS THE ANGLE FOR THE LOWER ORDER AND ?2 IS FOR THE HIGHER ORDER MODE). ...............................136FIGURE 6-3 THE CROSS-SECTION OF THE SUBWAY TUNNEL AND ITS EQUIVALENT RECTANGLE. ....................139FIGURE 6-4 ZERO-MEAN GAUSSIAN FIT OF THE PAS FOR THE SMALL TUNNEL (MULTIMODE WAVEGUIDEMODEL). ....................................................................................................................................................140FIGURE 6-5 ZERO-MEAN GAUSSIAN FIT OF THE PAS FOR DIFFERENT ZONES OF TWO TUNNEL SIZES(MULTIMODE WAVEGUIDE MODEL). ...........................................................................................................141FIGURE 6-6 COMPARISON OF PAS OBTAINED BY MULTIMODE AND RAY-TRACING FOR THE SMALL TUNNEL(WHOLE TUNNEL).......................................................................................................................................142xiFIGURE 6-7 AS VARIATION OVER DISTANCE FOR TWO TUNNEL SIZES (MULTIMODE WAVEGUIDE MODEL). ....143FIGURE 6-8 COMPARISON OF THREE METHODS FOR CHARACTERIZING AS IN A LARGE TUNNEL (RAY-TRACINGAND EXPERIMENT GRAPHS ARE OBTAINED FROM [128]).............................................................................144FIGURE 6-9 COMPARISON OF DIFFERENT METHODS FOR CHARACTERIZING AS IN A LARGE TUNNEL, INCLUDING10 M-50 M ZONE (RAY-TRACING AND EXPERIMENT GRAPHS ARE OBTAINED FROM [128])..........................145FIGURE 6-10 CORRELATION COEFFICIENT OF ULA ANTENNAS VS. THEIR SEPARATION FOR SMALL AND LARGETUNNELS ACROSS 10 M-500 M DISTANCE (MULTIMODE WAVEGUIDE MODEL). ...........................................148FIGURE 6-11 COMPARISON OF 4?4-MIMO CAPACITY VS. TX-RX DISTANCE FOR 4 ANTENNA SPACINGS(SNR=20 DB)............................................................................................................................................151FIGURE 6-12 4?4-MIMO CAPACITY CCDFS FOR SEVERAL ZONES (10 M- 500 M) AND TWO INTERELEMENTSEPARATIONS OF ?/2AND 6? (SNR=20 DB)...............................................................................................152FIGURE 7-1 MAP OF THE MYRA FALLS MINE IN BRITISH COLUMBIA, CANADA AND THE TRANSMITTER (TX)AND RECEIVER (RX) LOCATIONS. ...............................................................................................................162FIGURE 7-2 EXPERIMENTAL VALIDATION OF MULTIMODE MODEL IN A SHORT UNDERGROUND MINE TUNNEL;ANTENNAS ARE PLACED AT MEDIUM HEIGHT AND CLOSE TO THE CEILING. ................................................163FIGURE 7-3 D-MIMO AND C-MIMO SETUPS FOR AP-TO-MOBILE COMMUNICATIONS INSIDE AN EMPTYRECTANGULAR TUNNEL. ............................................................................................................................166FIGURE 7-4 RECEIVED POWER CDFS FOR DIFFERENT C-MIMO ACCESS-POINT CONFIGURATIONS. ...............170FIGURE 7-5 RECEIVED POWER CDFS FOR DIFFERENT D-MIMO SYSTEM SIZES. ............................................171FIGURE 7-6 CAPACITY CDFS FOR DIFFERENT C-MIMO ACCESS-POINT CONFIGURATIONS. ..........................172FIGURE 7-7 CAPACITY CDFS FOR DIFFERENT D-MIMO SYSTEM SIZES. ........................................................173FIGURE 7-8 CONDITION NUMBER CDFS FOR DIFFERENT D-MIMO SYSTEM SIZES.........................................174FIGURE 7-9 2?2 D-MIMO SYSTEM PERFORMANCE FOR SEVERAL SCENARIOS...............................................176FIGURE 7-10 2?2 D-MIMO POWER DISTRIBUTION FOR DIFFERENT CROSS-SECTIONAL LOCATIONS OF APS INTUNNELS WITH DIFFERENT LENGTHS..........................................................................................................177FIGURE 7-11 2?2 D-MIMO CAPACITY DISTRIBUTION FOR DIFFERENT CROSS-SECTIONAL LOCATIONS OF APS INTUNNELS WITH DIFFERENT LENGTHS..........................................................................................................177FIGURE 7-12 COMPARISON OF DIFFERENT SIZE D-MIMO SYSTEMS FOR SEVERAL TUNNEL LENGTHS WITH APLOCATIONS CLOSE TO THE SIDEWALL AND CEILING (1/5TH- 1/4TH OF THE WIDTH AND HEIGHT)................178xiiABBREVIATIONSAP Access PointAS Angular SpreadBS Base StationBWc Coherence BandwidthCAD Computer Aided DesignCCDF Complementary Cumulative Distribution FunctionCDF Cumulative Distribution FunctionCDMA Code Division Multiple AccessCIM Cascade Impedance MethodCIR Channel Impulse ResponseC-MIMO Collocated MIMOCN Condition NumberDAS Distributed Antenna SystemdBi deciBel isotropicD-MIMO Distributed MIMODOA Direction Of ArrivalDOD Direction Of DepartureDS Delay SpreadEM ElectromagneticFDTD Finite Difference Time DomainFM Frequency ModulationFSPL Free Space Path LossGA Genetic AlgorithmGO Geometry OpticsGPIB General Purpose Interface BusGSM-R Global Systsem for Mobile communications - RailwayGTD Geometrical Theory of DiffractionxiiiHVAC Heating, Ventilation and Air ConditioningIEEE Institute of Electrical and Electronics Engineersi.i.d. independent and identically distributedISI Intersymbol InterferenceISM Industrial, Scientific and MedicalLAN Local Area NetworkLF Low FrequencyLOS Line Of SightLTE Long Term EvolutionMATLAB MATrix LABoratoryMF Medium FrequencyMIMO Multiple-Input-Multiple-OutputMINER Mine Improvement and New Emergency ResponseNB Narrow BandNLOS Non Line Of SightPAS Power Azimuth SpectrumPCS Personal Communications ServicePDP Power Delay profilePEC Perfect Electric ConductorPED Personal Emergency DevicePL Path LossPLC Programmable Logic ControllerPNA Professional grade Network AnalyzerPSO Particle Swarm OptimizationRF Radio FrequencyRFID Radio Frequency IDentificationRMS Root Mean SquareRSL Radio Science LabRx ReceiverSBR Shooting Bouncing RaySCPI Standard Commands for Programmable InstrumentsxivSHF Super High FrequencySISO Single-Input-Single-OutputSNR Signal to Noise RatioSSM Segmental Statistical MethodTE Transverse ElectricTM Transverse MagneticTL Transmission LineTOA Time Of ArrivalTTA Through The AirTTE Through The EarthTTW Through The WireTx TransmitterUBC University of British ColumbiaUHF Ultra High FrequencyUMTS Universal Mobile Telecommunications SystemULA Uniform Linear ArrayUTD Uniform Theory of DiffractionUWB Ultra Wide BandVHF Very High FrequencyVLF Very Low FrequencyVNA Vector Network AnalyzerWI Wireless InSiteWiMAX World wide interoperability for Microwave AccessWLAN Wireless Local Area NetworkWPAN Wireless Personal Area NetworkxvACKNOWLEDGEMENTSPursuing my Ph.D. was one of the most precious experiences I have ever had in mylife. This achievement would not have been possible without the help, support and inspirationof many people around me.First of all, I would like to express my sincere appreciation and gratitude to mysupervisor. I had the honour to work with Prof. David G. Michelson who is a successful, andwell-known professor in both experimental and theoretical research. His valuable knowledge,experience, comments, and suggestions have improved different aspects of my PhD program.Throughout my PhD, he has been a great role model for conducting research, writing skills,and technical presentations. Many thanks to him.My special thank to Prof. Sima Noghanian for her time and effort she devoted duringmy four-month visit to University of North Dakota under the Michael Smith Foreign StudySupplements Program (i.e., CGS-MSFSS, granted by NSERC). I learnt a lot from our fruitfultechnical discussions.Thanks to my kind colleagues Robert White, Pooyan Abouzar, Wadah Muneer,Shahzad Bashir, Carol Lee, James Chuang, Anthony Liou, and Kyle Sivertsen for their kindsupport and help for different research activities, particularly in conducting MIMOmeasurement campaigns in a service tunnel at UBC and in Myra Falls located in VancouverIsland, BC. It is a privilege to be surrounded by such kind people who are always available tohelp; I appreciate all of them.xviI would like to thank National Sciences and Engineering Research Council of Canada(NSERC), Western Economic Diversification Canada (WD) and Modular Mining, andBritish Columbia Innovation Council (BCIC) for supporting my Ph.D. research through anNSERC Alexander Graham Bell Canada Graduate Scholarship, a Graduate ResearchAssistantship and a BCIC Innovation Scholars Award, respectively. Such supporttremendously impacted my Ph.D. life. They brought me great inspiration and encouragementto continue challenging journey of Ph.D. program with strength and confidence. Theequipment provided to our lab under grants from the Canadian Foundation for Innovation(CFI), WD and NSERC was essential to completion of the experimental portions of thiswork.Finally, I would like to thank UBC Building Operations for providing us with accessto the service tunnel underneath the Woodward Instructional Resources Centre and NystarMining for providing us with access to their Cu-Pb-Zn mine near Myra Falls on VancouverIsland for the purpose of collecting measurement data. Such access was also essential tocompletion of the experimental portions of this work.xviiDEDICATIONTo my dear parents, sister,and toall brave women and men who have sacrificed their lives and families to makethe world a better place for others......toRachel Corrie (peace activist)Sattar Beheshti (blogger),Nasrin Sotoodeh (human rights lawyer),Bradley Manning (soldier),??.1CHAPTER 1: INTRODUCTION AND OVERVIEWIn recent years, the underground mining community has begun to embrace short-range wireless communications technology as a key part of their strategy for enhancing thesafety and productivity of their operations. Multiple-input-multiple-output (MIMO) antennaarray technology has recently emerged as an important technology for boosting the capacityof wireless networks in conventional residential, commercial and industrial environmentsthough exploitation of multipath propagation. However, wireless propagation in tunnels andunderground mines is quite different from that in conventional surface environments. Here,we seek to assist those seeking to deploy MIMO-based wireless systems in undergroundmines by: 1) characterizing wireless propagation in straight sections of drifts or tunnelstypical of those found in modern hard rock mines, 2) assessing the manner in which theconfiguration and placement of both conventional and distributed MIMO antenna arrays andtransmitter-receiver separation affects the performance of such systems and 3) validating ourresults by employing alternative methods for characterizing wireless propagation (andMIMO-based wireless system performance) in underground environments.In 2008, the mining industry in British Columbia generated gross revenues of $8.4 billion(up from $6.9 billion in 2007) with net earnings of $3.2 billion (up from $ 1.2 billion in2007). The mining industry in British Columbia directly employs some 7,600 people andindirectly employs another 21,000. The province is also home to several companies,including Modular Mining Systems Canada, Nautilus Automation and WencoInternational Mining Systems, which provide mine communications and automationsystems to clients around the world. Accordingly, methods for making underground minecommunications more reliable and less expensive is of obvious strategic importance tothe provincial economy [1].21.1 Wireless in Underground Mines1.1.1 Legacy SystemsAlthough early efforts to deploy wireless technology in underground mines andtransportation tunnels date back to the 1920?s [2],[3] much of this work was experimentaland lacked theoretical support. Moreover, most early work was conducted at relatively lowfrequencies and achieved only limited success due to the extremely high path losses that wereencountered [4]. In the late 1960?s and early 1970?s, safety concerns prompted governmentregulators and safety boards in both Europe and North America to encouraging the miningindustry to improve communications with underground workers by deploying wirelesssystems based upon VHF-FM portable radios and leaky feeder distribution systems [3]. AtVHF, wireless coverage in tunnels and drifts in underground mines and transportationtunnels is very poor because these structures function as waveguides operated below cutoff.Radiating coaxial cable, popularly known as leaky feeder, was developed during the 1950?sand 60?s in order to extend the coverage of VHF wireless communication systems to therelatively short underground transportation tunnels often found in major urban centres [5].Underground mines presented a special challenge because they are typically far greater inextent, have much more complicated geometries, and evolve more rapidly over time thantransportation tunnels.Although leaky feeder based distribution systems have played an important role inunderground wireless communications for almost four decades, their limitations inunderground mining environments have become apparent in recent years. First, they are fixedinfrastructure that is susceptible to damage from blasting if placed too close to the face, i.e.,the blind end of a drift or closed tunnel that protrudes into an ore body and the place where3most of the activity in a mine takes place. That wireless coverage tends to be weakest whereit is needed most is a serious limitation. Second, it is time consuming and expensive todeploy leaky feeder, reconfigure it, or extend it as the face is pushed forward into the orebody (at rates between 2 m and 10 m per day, depending on the mine type). Finally, currentlydeployed leaky feeder distribution systems are not compatible with recently developedwireless standards and technologies which offer much higher performance and greatercapability, e.g., integrated voice and data.1.1.2 Modern SystemsMany of the limitations of VHF leaky feeder distribution systems can be avoided bymoving to higher frequencies where signal attenuation is vastly reduced. Althoughcommercial radios capable of operating at frequencies above 850 MHz have been availablesince the mid-1980?s, interest in replacing existing leaky feeder distribution systems did notmaterialize immediately (Figure 1-1). By the late 1990?s, growing interest in mineautomation and tele-operation of mining equipment and the advent of standards-based digitalwireless communication systems that operate above 800 MHz finally rendered leaky feederdistribution systems obsolete. Operating at higher frequencies eliminates the need forexpensive leaky feeder cable infrastructure and maintenance, permits rapid redeployment andreconfiguration of the system as required, and allows effective wireless communication to becarried out at the face.  The new systems also permit voice, data and video communicationsto be provisioned using a common infrastructure.During the late 1990?s and early 2000?s, several mining companies, including IncoLtd., evaluated the performance of various UHF wireless communications systems including1.8 GHz PCS CDMA cellular and ISM 915 and ISM 2450 Wireless LAN in underground4mine environments. The mining industry has recently started investing in future-orientedtechnologies such as machine-to-machine (M2M) solutions, offering a high degree ofoptimization potential in deeper underground mines. This requires a solid wirelesscommunication platform to simultaneously handle various services and devices such as,hand-held devices, remotely-controlled machines, cameras, dispatches, RFIDs, toxic gas-detectors, thermal, humidity, and geological sensors, etc.Figure 1-1 Technology push into mining underground mines over time.Among different standardized products, IEEE 802.11-based wireless LAN-basedsystems that operate in the ISM 2450 band have attracted the most interest because they offeran effective combination of economy and flexibility in deployment and configuration.However, performance of such systems is uncertain because most of them are based onmultiple-antenna systems which have not been studied for applications in undergroundmines. While performance degradation due to channel impairments is troublesome whenusers are accessing voice or information services, it can be intolerable in mine automation or5tele-operation applications where response times are critical. As a result, MIMO-basedwireless systems, well known for offering high performance in conventional environments,need to be modeled and carefully designed in underground mines.1.2 Characterization of MIMO-Based SystemsIn this section, we recall two key parameters of MIMO systems: the first one ischannel capacity in bit/sec/Hz, a performance parameter, while the second one is angularspread, an environmental parameter.1.2.1 MIMO Channel CapacityScattering by objects in the environment leads to multiple transmission paths betweenthe transmitter and receiver. While this raises the possibility of increasing link capacity byutilizing each path as an independent channel, such paths are often so closely spaced in anglethat one cannot distinguish between them using simple beamforming. A more sophisticatedapproach is to use space-time coding to distribute the data stream over the NTx transmittingantennas and recover the stream by suitably combining the signals received by the NRxreceiving antennas. The resulting multiplexing or capacity gain, is achieved without requiringeither additional power or additional bandwidth [6].Capacity, which is a fundamental property of a MIMO-based system, may bepredicted as follows. First, consider a wireless communications system that uses only onetransmission path to send data. Shannon?s Law gives the maximum capacity C1 of the link inbit/s/Hz as:? ?1 2log 1C ?? ? (1)6where ? is the signal-to-noise ratio (SNR) at the receiver input. The capacity C (in bit/s/Hz)of a MIMO-based system with NTx transmitting antennas and NRx receiving antennas is givenby [6]:*2log det RxN nor norTxSNRC I H HN? ?? ?? ?? ?? ?? ?? ?b/sec/Hz (2)where ? denotes the transpose-conjugate, H is the NRx ? NTx channel matrix and we haveassumed that the NTx sources have equal power and are uncorrelated.In a rich multipath environment where signal fading observed by individual receivingantennas is uncorrelated, the channel matrix has full-rank and the increase in capacity isproportional to the minimum of the number of transmitting and receiving antennas. Becausethe multiple transmission paths fade independently of each other in such cases, this approachalso increases overall link reliability. If the subchannels are highly correlated then thechannel matrix will be rank deficient and hence no capacity gain is achievable. This mayoccur when the angular spread of incoming signals at the receiver is very low. It is apparentthat full appreciation of the strengths and limitations of alternative MIMO schemes requiresthat their performance be assessed [7].1.2.2 Angular SpreadThe main focus of characterization of single-input single-output (SISO) systems is thepathloss and delay spread. For characterization of MIMO channels, however, the spatialdomain becomes equally important as the temporal domain. Similar to power-delay-profile(PDP), power-azimuth-spectrum (PAS) has been defined, which determines the spatialdistribution of the received power over the azimuth domain. Consequently, angular-spread7(AS) is defined as the standard deviation of PAS, being equivalent to the root-mean-square(RMS) delay-spread of PDP [8]. AS is found to govern many properties of MIMO-basedsystem metrics, e.g., singular value and capacity distributions [9]. In addition, parameter thatcharacterizes space selective fading is the coherence-distance (DC) that is inverselyproportional to the AS. The DC is the spatial separation for which the autocorrelationcoefficient of the spatial fading drops to 0.7 [10].In conventional indoor and macrocell environments, the PAS at the mobile unit tendsto be broad so fading on adjacent antenna elements tends to be uncorrelated for relativelysmall element separations (smaller DC). Linear confined spaces such as tunnels and drifts inunderground mines tend to function as overmoded waveguides at radio frequencies. As aresult, the PAS in such environments is expected to be considerably narrower than inconventional environments and the performance of MIMO wireless systems, is likely to bereduced. However, no experimental or theoretical study has been performed to predict orestimate angle-of-arrival distribution.Consequently, we decided to theoretically study angular properties of the tunnels.Several theoretical approaches, such as single-mode waveguide model and geometrical-optical models (i.e., are explained in great detail in Chapter 2) [11]-[16] were developed topredict propagation in underground tunnels at UHF and above during the 1970?s. For ourstudy, we chose multimode waveguide model which has been recently developed. Thismodel is the advanced version of the single-mode waveguide model and is able to accuratelymodel both near-field and far-field of tunnels.81.3 Previous Efforts to Design Antenna Configuration in IndoorEnvironmentsConsidering different antenna properties, such as radiation pattern, polarization andarray configuration and spacing (in case of MIMO systems) introduces the flexibility in aneffective design of communication systems. Here, the main antenna properties that are takeninto account in system designs and deployments for indoor areas are discussed.1.3.1 Antenna Radiation PatternFor antenna design and deployment of SISO systems in outdoor and indoorenvironments, the main focus is coverage and interference. Antenna radiation pattern (omni-directional or directional) plays a key role in coverage and interference. Use of directionalantenna is well-established for outdoor and indoor environments to increase coverage andreduce interference [17]. However, the disadvantage is that the performance of directionalantenna is highly dependent on the antenna orientation while optimal orientation itself isdependent on the layout of the environment and propagation scenario [17], [18].In [17], the performance of omni-directional discone and directional patch antennaswas compared in the presence of internal obstacles and external interference in an indoorenvironment. The general suggestion of this study is to illuminate distinct regions of the floorarea and away from each other with different directional APs. This guarantees sufficientlyhigh signal reception across the entire area while minimizing the mutual interference betweenAPs and mitigating external interference, which leads to improvement in overall systemperformance. However, signal coverage is compromised by this improvement and thusadditional APs may be required at the expense of increased deployment complexity.91.3.2 Antenna PolarizationUnlike outdoor environments where different antenna polarizations perform quitedifferently (mainly due to the ground effect), in conventional indoor environments, such asoffices and labs they behave similarly. However, in hallways the results are a bit different.Researchers at Stanford University have conducted narrowband measurements at 1.95 GHzin a hallway (LOS paths) with laboratories (NLOS paths) on both sides of the hallway[19],[20]. The Tx was located in one end of the hallway and Rx was moved along thehallway and inside the labs. They showed that horizontally polarized waves attenuate morequickly than vertically polarized waves in the hallway but they both attenuate at about thesame rate in the labs. It has been attributed to the Brewster angle phenomena which happenin interaction of horizontally polarized waves with dielectric sidewalls (which unlikeconductive floor and ceiling behave differently for different polarizations).The effect of antenna polarization has been studied for SISO systems in tunnels too.The main result of the studies is that in empty tunnels with horizontal aspect ratio (i.e., widthis larger than height), attenuation of the horizontal polarization is lower. However, theirperformance for MIMO systems and in presence of infrastructure is uncertain and needs to bestudied.1.3.3 Antenna Array Configuration and SpacingIn MIMO systems, in addition to coverage and interference, the correlation amongantenna elements is also important. This correlation is also influenced by antenna properties,such as interelement spacing, array orientation, antenna radiation pattern, and polarization aswell as angular spread of the wireless channel. Because conventional indoor environments10such as offices are often rich in terms of scattering, angular spread is sufficiently large.Therefore, different array configurations, such as UCA and ULA with both polarizationsshow suitable performance. For indoor offices, ?/2 interelement separation betweenneighboring antenna elements show sufficient level of decorrelation between antennaelements, however, in hallways the spacing should be larger than ? [19],[20].Figure 1-2 Different types of MIMO antenna configurations, such as ULA and UCA for indoorenvironments.A research group at the University of Lille in France has conducted several studies of4?4-MIMO wireless channels in transportation tunnels used for railways, roadways and/orsubways [21]-[24]. These studies have focused on the 900 MHz band for GSM-R (a variantof GSM designed for use by high speed railways) applications. Several ULA orientationswere compared. The Tx antennas were fixed and placed on the station platform withfollowing array orientations: 1) parallel to the tunnel axis, 2) diagonal (along a line at anangle of 30o from the centreline), and 3) perpendicular to the tunnel axis. Due to safety andclearance issues, mobile antennas cannot be installed on the train roof, therefore, the mobileRx was installed on the train?s windshield. In order to have sufficient degree of fadingdecorrelation among MIMO antennas, antenna spacing at Tx was more than 3? and1.4? at11Rx. Based on this study, the MIMO performance found to be significantly sensitive to the Txarray orientation, which is due to the low AS caused by the waveguide effect inside thetunnel. The best performing orientations were diagonal and the one aligned perpendicular tothe tunnel axis.Figure 1-3 Fixed Tx ULA placed on the platform, Mobile Rx ULA installed on the train windshield[21].1.4 Implications for MIMO Antenna Configuration Design andDeployment in Underground MinesWhile previous studies of MIMO modeling and configuration design in otherconfined spaces have yielded useful insights, they cannot replace studies conducted inunderground mines. As a result, mine communication engineers have had little to guide themas they seek to design and deploy MIMO-based wireless systems underground. Compared totransportation tunnels (the most similar to mine tunnels), underground mine tunnels andgalleries are considerably narrower which may lead to higher modal cut off frequencies andlower angular spread that will limit MIMO performance. In addition, mine tunnels havebranches, considerably rougher walls, and irregular geometry, which cause much more12diffuse scatter. As a result, further study is required in order to assess the relative magnitudeof these effects and to determine the MIMO performance that can be achieved.1.5 Objectives of This WorkWe have sought to contribute to the safe and efficient operation of undergroundmines and transportation systems and overcome the limitations of past work by: 1)developing a MIMO measurement system to experimentally assess the MIMO performancefor various deployment scenarios in underground mine environments, 2) using threecomplementary modelling and prediction tools: i) field measurements using a custom-designed channel sounder in both a building service tunnel at the University of BritishColumbia and an underground lead-zinc mine at Myra Falls, BC, ii) simulations based uponray-tracing and iii) theoretical models based upon waveguide mode expansion to assess andcompare strategies for effective configuration and deployment of MIMO antennas inunderground mines, 3) experimentally validating a theoretical model for tunnels (multimodewaveguide model), which has not been validated for short mine tunnels, 4) characterizingangular-spread and MIMO channel capacity based on this theoretical model and5) developing techniques to assess and optimize the performance of distributed MIMOsystems based on the multimode waveguide model.1.6 Organization of the ThesisThe organization of the thesis is as follows: In Chapter 2, we present a comprehensiveliterature survey on wireless developments and propagation modeling in underground minesfrom 1920-2012 is presented. In Chapter 3, we describe our MIMO measurement setup anddata acquisition software. In Chapter 4, we present and compare the results of MIMO13simulations and measurement campaign conducted in a service tunnel at UBC. In Chapter 5,we present the results of a MIMO measurement campaign performed in the underground Cu-Pb-Zn mine at Myra Falls, BC operated by Nystar Mining. In Chapter 6, we present theresults of characterizing MIMO wireless channels in an underground mine using a multimodewaveguide model. In Chapter 7, we present a novel method for optimizing the placement ofdistributed MIMO antenna elements in underground mines using multimode waveguidemodelling and particle swarm optimization. Finally, in Chapter 8, we present the conclusionsof this study and suggest possible next steps.14CHAPTER 2: A SURVEY OF WIRELESSCOMMUNICATIONS AND PROPAGATIONMODELING IN UNDERGROUND MINES2.1 IntroductionThe mining industry plays a vital role in the global economy. The current estimatedmarket capitalization of global mining companies is about $962 billion [25],[26]. A largeportion of these operations are underground and involve specialized equipment andprocesses. Communication systems play an increasingly important role in ensuring personnelsafety and optimizing the mining process. The estimated size of underground miningequipment market alone is currently about $45 billion [27], a small but important portion ofwhich is allocated communications systems.Although interest in deploying wireless communication systems in undergroundmines dates back to the 1920's [28],[29] the first wide deployment didn?t take place until theearly 1970's when the mining industry began to deploy very-high-frequency (VHF) radiosand leaky feeder distribution systems [30]-[34]. The modern era of undergroundcommunications began in the early 2000's as the mining industry sought to take advantage ofconsiderable advances in ultra-high-frequency (UHF) technology, especially cellular phones,wireless-local-area-network (WLAN), UWB and radio-frequency-identification (RFID).Although the mining industry is generally conservative and reluctant to invest in costly newtechnologies, high profile accidents often prompted regulators to require that the mining (and15mining communications) industry devote increasing attention to safety and safetycommunications [35]. Recent interest in deploying next generation wireless communicationstechnology in underground mines has stemmed from: (1) recent advances in short-rangewireless communications technology and commercial-off-the-shelf WLAN, wireless-personal-area-network (WPAN), UWB, RFID, radar devices, and (2) the potential to increasemine efficiency and productivity through more effective voice communications, better accessto management information systems and automated dispatch [36],[37].In an underground mine, there are three possible mechanisms for communicationsignaling: through-the-earth (TTE) at extremely-low-frequency (ELF)/very-low-frequency(VLF)/low-frequency (LF) bands, through-the-wire (TTW) at medium-frequency(MF)/VHF/lower-UHF (e.g., leaky feeders) and through-the-air (TTA) at upper-UHF/super-high-frequency (SHF) [38]. Each has been developed for different applications and eachrequires specified propagation channel modeling and design. Most of the recent wirelesssystems fall under the TTA category and also seem to be promising wireless technology forfuture applications. Accordingly, the main focus of this survey is on methods forcharacterization of the TTA wireless channels at UHF-band; TTE and TTW are only brieflyconsidered.The need to understand and characterize wireless channels has been recognized sincethe earliest days of wireless communications. The objective of channel characterization ormodeling is to capture understanding of the manner in which the propagation environmentimpairs and distorts wireless signals in a form useful in the design, test and simulation ofwireless systems. Subway tunnels lack the rough and tilted walls that characterize16underground mines. However, because their propagation characteristics show somesimilarities to those of underground mines, they have been considered here as well.For decades, researchers have recognized and studied the differences betweenwireless propagation in tunnels and underground mines and surface environments. Valuabletheoretical and experimental contributions have been made by several groups includingJ. Wait et al., S. F. Mahmoud, A. E. Goddard et al., A. G. Emslie et al., P. Delogne, Y. P.Zhang et al., M. Lienard et al., and C. Despins et al. Two non-recent reviews, one from 1978[31] and one from 1991 [34], are about early stage wired/wireless communicationtechnologies such as different types of phones, pagers, leaky feeders and TTEcommunications. In 2009, the Canada Center for Mineral and Energy Technology(CANMET) reviewed the current state of wireless communications technology forunderground mines including products manufactured by key suppliers, their specifications,limitations and advantages [39]. Products by key companies, such as Becker MiningSystems, Mine Radio Systems, MineSite Technology, MineCom, Tunnel Radio of Americaand Varis, are evaluated in this study. In addition, for documents involving safety andpermissible designs for electronic communications systems, the National Institute forOccupational Safety and Health (NIOSH) has provided online resources such as collectionsof past and current mine communications publications, tutorials and workshops [40]. Inanother recent survey [41], both past and current communication systems in wired/wirelessforms are introduced and the significance of each is briefly discussed.Despite of all of this past effort, there is no comprehensive survey to date ofunderground communications that not only introduces the technologies and their significancebut also reviews the propagation channel models developed for underground tunnels and17mines. This can be a barrier for those who want to enter into this field or would like to knowmore about the subject matter.In this chapter, we aim to present a comprehensive survey of wireless propagation intunnels and underground mines with a focus on current wireless channel modeling,technologies and applications. Our objective is to put previous work in perspective, identifytrends and gaps, and summarize accomplishments and opportunities. In Sec. 2, we begin thesurvey with a brief review of the basic wireless propagation terminology. In Sec. 3, wepresent a brief history of wireless communications in underground environments. In Sec. 4,we show how the related numerical and analytical models have evolved over time. In Sec. 5,we consider measurement-based models. In Sec. 6, we summarize the practical implicationsfor wireless system design based on significant contributions of several researchers. Finally,in Sec. 7, we present the conclusions of this chapter.2.2 Wireless Propagation TerminologyIn this section, some technical terms that will be required throughout the dissertationare addressed and explained. These terms will be described by explaining how a transmitsignal undergoes pathloss and fading before reaching a receiver. Most of material presentedhere has been extracted from a comprehensive survey on propagation models for mobilecommunications in [42] as well as a book chapter [43].As a wireless signal traverses the path from a transmitter to a receiver, it experiencesdifferent propagation phenomena such as reflection, diffraction, scattering and refraction(Figure 2-1). Reflection occurs when the electromagnetic (EM) wave is incident upon asmooth surface, whose dimensions are large compared with the signal wavelength.18Diffraction is a propagation scenario in which an object whose dimension is larger than thesignal wavelength and which has sharp edges obstructs a path between transmitter andreceiver and causes new secondary waves to be generated. Scattering occurs when incomingsignal is incident upon an object whose size is of the order of the wavelength of the signal orless. Refraction is the change in direction of a wave due to a change in its velocity whiletraveling between media with different refractive indexes.Figure 2-1 Wireless propagation phenomenaAs a result of interaction of signal with the surrounding area, replicas of the signalmay take multiple paths from the transmitter to the receiver. Because the replicas reach thereceiver after different delays, the signal experiences time dispersion (quantified by delayspread) and because they also arrive from different directions, the signal experiences angulardispersion (quantified by angular spread) [43]. If either the scatterers or one of the terminals(Tx or Rx) moves, rapid changes in the phase relationship between multipath componentscan cause the signal to fade randomly, i.e., fading. Such variation in received signal strengthover time is equivalent to frequency dispersion (quantified by Doppler spread). Fading can be19categorized into two main types: small-scale fading and large-scale fading which are shownin Figure 2-2.Figure 2-2 Large-scale and small-scale fadingsSmall-scale fading is due to small changes in position (as small as half wavelength) orto changes in the environment (surrounding objects, people crossing the line of sight betweentransmitter and receiver, opening or closing of doors, etc.). Small-scale fading models,therefore characterize the rapid fluctuations of the receiver signal strength over very shorttravel distances (few wavelengths).Large-scale fading is due to motion in a large area, and can be characterized by thedistance between transmitter and receiver. Large-scale fading models therefore predict themean signal strength for an arbitrary transmitter-receiver (Tx-Rx) distance are useful forestimating the radio coverage. Pathloss and pathloss exponent (or distance exponent, power-distance-factor) are terms used in large-scale models for indoor and outdoor environment.20Pathloss, the most fundamental measure of channel quality, is the attenuation of thetransmitted signal as it propagates. In decibels, pathloss, PL is defined as:Tx Tx Rx RxPL P G G P? ? ? ? (1)where PTx and PRx are the time-averaged power levels (in dBm) at the output of the Tx andthe input of the Rx, respectively, and GTx and GRx are the gains (in dBi) of the transmittingand receiving antennas. The relationship between pathloss and the distance, d, between theTx and Rx follows a power-law relation and can be described by [43]:0 100( ) 10log dPL d PL n Xd ?? ? ? (2)where PL0 is the value of pathloss (in dB) at the reference distance d0, n is the distanceexponent and X? is a zero-mean Gaussian random variable with standard deviation ?. Therandom variable X accounts for the location variability or shadow fading that is generallyattributed to differences in the degree to which the path is obstructed at different pointsthroughout the coverage area.The pathloss exponent is a measure that yields to what power of separation the signalpower in the profile decays. It can be determined by applying regression analysis to the largefading component of a received signal data file. It is in the range of 2 (for free space) to 4(for the case of specular reflection from the ground surface). In some environments, such asbuildings and other indoor environments, the pathloss exponent can reach values in the rangeof 4 to 6. On the other hand, a hallway or a tunnel may act as a waveguide, resulting in apathloss exponent less than 2. Most of pathloss models have one or several breakpoints21which distinguish areas where radio wave experiences different pathloss exponents(Figure 2-3).As an example, two-ray model (or flat-earth model) which accounts for the specularreflection from the earth surface has one breakpoint. Before that pathloss exponent is 2 andafter that due to contribution of the reflected ray from the ground, pathloss exponent changesto 4.Figure 2-3 A pathloss model with three breakpointsTunnel pathloss model has also a breakpoint which separates far-field (or far zone)and near-field (or near zone) regions. The breakpoint location in a tunnel depends on thelargest cross-sectional dimension (width or height) of the tunnel relative to the signalwavelength. It should be noted that far-field and near-field definitions for propagationmodels in tunnels are not the same as far-field and near-field of an antenna.The near-field region of a straight tunnel is the region before the breakpoint wheresignal fluctuation is significant because of the reflections and multipath components comingfrom different directions, which are comparable to the line-of-sight path. In this region, the22pathloss exponent is closer to that of indoor environment (n ? 2). On the other hand, the far-field region of a straight tunnel is the region after the breakpoint where all the paths arereceived at the Rx from almost the same angle as the direct path, and therefore the receivedsignal is well established and the fluctuation of the signal is not significant. In this region,pathloss exponent is less than that of free space pathloss, i.e., n=2, and so-called waveguidepropagation occurs.The attenuation constant describes the attenuation of an EM wave propagatingthrough a dielectric medium per unit distance from the source. It is the real part of thepropagation constant, measured in Nepers per metre (Np/m) and accounts for attenuation dueto propagation in a lossy environment. Attenuation constant of vacuum equals to zerobecause it is a lossless medium.  Assuming a transverse-EM (TEM, i.e., E and H are bothperpendicular to direction of propagation) plane wave propagating in the z direction can berepresented using the phasor expression:0( , ) zE z E e ?? ?? (3)where ? is the radian frequency and ? is the complex propagation constant given as:? ? ? ? ? ? (4)??????? j??where ? (Np /m) is the attenuation constant and ? (rad /m) denotes the phase constant.2.3 The Evolution of Wireless Communications in Tunnels andUnderground MinesIn this section, the evolution of wireless communications in underground mines isdiscussed in terms of technologies and applications. Both reveal that the initial motivation for23underground mine communications was to increase the safety of miners by implementingman-to-man communications. As underground mine communications have evolved, man-tomachine and machine-to-machine communications have been implemented to meetefficiency and productivity objectives.2.3.1 Through-The-Earth CommunicationsInterest in wireless communications for underground mine dates back to the 1920?swhen the earliest pioneers of radio were interested in the possibilities of TTE wirelesstransmission. N. Tesla suggested to use ELF signals, and the earth as a transmitting mediumto send messages across the world in 1899 [44]. This continued until the late 1940?s whentechniques such as carrier-current radios and TTE signalling were commercially offered bythe U.S. Bureau of Mines for ordinary communications and for emergency operations inmines [28],[29],[34]. TTE communications in mines use huge antennas to transfer ELF orVLF signals through solid rock from the surface into the underground mine. In late 1940?s,due to limitations such as low data rate and bulky mobile equipment, early studies of wirelesscommunications in tunnels were terminated [45],[46].Recent mine regulations have renewed interest in TTE communications because itoffers a wider coverage inside the mine compared to modern wireless systems. There areapparent advantages to modern wireless systems in underground tunnels and mines, but theycould be quite vulnerable when a major disaster occurs. Disasters such as explosion,flooding, rock burst, or severe roof fall, may damage the relay system or block airways. TTEcommunications has been proven to be suitable for emergency communications because itaccesses every part of the mine by propagating through the rock and requires no cablingbetween the surface and underground [47]. Two-way communication systems are preferred24over one-way systems because in most emergency cases, it is essential for escaping ortrapped miners to relay valuable information to the surface.Until several fatal incidents occurred in the United States in 2006, the number ofmining disasters had been following a decreasing trend. The Mine Improvement and NewEmergency Response (MINER) Act of 2006 requires that mine operators install wirelesstwo-way communications and tracking systems that will connect surface rescuers to theunderground workers [48]. Two commonly used wireless solutions for emergency cases aretext messaging based on TTE and tracker tagging.Personal-emergency-device (PED) is an emergency warning system based on TTEtechnology [49], which uses VLF/ULF signals to transmit text messages (Figure 2-4).Initially, this product had one-way communication capability, but recent versions are capableof two-way communication via text messaging [49].Figure 2-4 Through-the-earth communications252.3.2 Through-The-Wire CommunicationsIn the early history of through-the-wire communications in tunnels and undergroundmines, implementation of communication systems was based on experimental observationswithout any theoretical insights or empirical modeling attempts. People working inunderground mines found that low frequencies on the order of 10 MHz (cutoff frequency offundamental modes of most tunnels) could cover distances of less than 30 m in an emptymine [12]. However, they also observed that conductors such as electrical cables, pipes andetc., running in most mines, enhance EM propagation with low attenuation, and thereforeincrease the range [44]. This fact was not immediately understood by experimenters, but itresulted in development of monofilar technique at the end of 1960?s. Monofilar systembecame an introduction for leaky feeder systems which were widely used thereafter.In general, TTW signals can travel over coaxial, twisted pair, trolley, leaky feeder,and fibre optics from the surface or inside the mine and reach the mobile equipment. Becauseone side of the system is wired and the other is wireless, it is also called a hybrid or semi-wireless system. During the 1950?s and 1960?s, leaky feeder systems and other distributedantenna systems were developed in order to extend the coverage of VHF wirelesscommunication systems to the relatively short underground transportation tunnels oftenfound in major urban centres and for providing public safety [33]. In the late 1960?s when thesafety concerns prompted government regulators and safety boards in both Europe and NorthAmerica to encourage the mining industry to improve communications with undergroundworkers by deploying wireless systems based upon VHF-FM portable radios and leakyfeeder distribution systems [34]. Leaky feeder is the most well-known TTW-basedcommunication system in underground mines. The cable is called ?leaky? as it has gaps or26slots in its outer sheath, allowing signal to leak into or out of the cable along its entire length(Figure 2-5). Because of this leakage of signal, line amplifiers must be inserted at regularintervals, typically every 350 to 500 metres.  Key disadvantages of leaky feeder system aredifficult maintenance, fixed infrastructure, limited capacity and low coverage near the face,i.e., the region of the mine where ore is extracted [34].Figure 2-5 Leaky feeder cable2.3.3 Through-The-Air CommunicationsTTA is another wireless system for communications in underground mines. It iscapable of offering various applications such as two-way voice and data communications,tracking miners and equipment, remote control and sensing, video surveillance and etc.In early 2000's, advances in short-range digital communications to cover hundreds ofmetres motivated the mining industry to consider WLAN off-the-shelf products to supportshort-range applications in underground. In the late 2000's, the mining industry was attractedto low data rate technologies such as ZigBee, active-RFID (tens of metres), passive-RFID(about a metre) and high data rate systems, such as UWB systems, because they offer short-range, low power and positioning capabilities. These technologies can support variousapplications such as dispatch and sensor networks. These applications can be implemented27based on WLAN backbone. So far, WLAN mesh networks which are redundant, self-learning and self-healing seems to be the most reliable wireless systems. If any part of thenetwork is destroyed, the remainder continues to function, and therefore it is especiallydesirable in a dynamic environment where link failures are frequent as in the mine galleries[37],[49]. One of the attractive wireless applications is tracking which can be implementedbased on RFID technology using WLAN, fibre optics or leaky feeder backbone (Figure 2-6).This tracking system provides the ability of real-time monitoring the location of personnel,vehicles and equipment underground. Mining equipment such as vehicles, containers, drillsand other valuable mobile ore production equipment are constantly moving through largeunderground areas. Because the equipment does not necessarily follow a pre-defined trackand is spread throughout the mine, it is difficult to locate particular assets that are needed inreal-time [50],[51].Figure 2-6 Tracking system in an underground mine [52].28A typical RFID-based tracker system is shown in Figure 2-6. This system consists of:(1) active tags to identify personnel/vehicles/assets or store data and histories, (2) tag readersto exchange information with the server and tags, (3) antennas to connect tags and tag readersand provide triangulation information for location finding, (4) a server computer system forcontrol and monitoring, and (5) backbone system which can be fibre optics or leaky feeder toconnect tag readers to the server [52].Another important application of short-range wireless is remote control and sensing.Some of commonly deployed control applications of wireless communication are real-timeremote equipment diagnostics, remote monitoring, remote programmable-logic-controller(PLC) programming, etc. As an example, a PLC in local control station can wirelesslycommunicate with the remote automation and sensor devices (such as pull cords, beltmisalignments and tilt switches or motion sensors) along a conveyor in a mine site.Before employing the aforementioned wireless technologies in tunnels andunderground mines, careful characterization of the wireless propagation in terms ofparameters such as pathloss, delay spread and angular spread, etc. is required. This is becausewireless propagation in tunnels and underground mines is significantly different fromconventional indoor and outdoor environments, and therefore existing channel modelsdeveloped for conventional surface environments are not applicable. Consequently, it will benecessary to develop new channel models that capture the nature of the relevant impairmentsand their dependence on the new environment.A good channel model is abstract, simple, and focuses on those aspects of the channelthat affect the performance of a system of interest and ignore the rest.  Over-engineering thecommunications links is needlessly expensive and under-engineering them leads to either29insufficient reliability or capacity. Propagation and channel modeling facilitates efficientdesign and system deployment by answering questions such as ?What channel impairmentsdo we need to mitigate? or ?What is the optimum frequency, antennaplacement/configuration and range??2.4 Propagation Analysis and ModelingIn this section, we focus on the different mechanisms by which wireless signalspropagate in underground environments. First, TTE (through the earth) and TTW (throughthe wire) propagation will be briefly explained. Then, as the main focus of this survey is TTA(through the air) communications, we will elaborate more on analytical modeling for thismethod. As it will be seen, TTE method is similar to geophysical probing, TTW method canbe analyzed similar to transmission lines and TTA in underground tunnels can be consideredas waveguide propagation.2.4.1 Propagation Through-The-EarthIn TTE underground communications, the antenna and propagation mechanism aresimilar to subsurface communications and geophysical probing of the Earth. Bothapplications require transmission of electromagnetic waves through the earth, and both facethe problem of high attenuation. To penetrate the earth to depths of 100 m or more, it isnecessary to employ low frequencies (ELF, VLF) [53]. For TTE communications, anextensive loop antenna or a ferrite rod antenna on the order of few kilometers is often used tosend/receive the magnetic waves through the rocks (Figure 2-7). The choice of magneticfields for TTE communications might be due to the fact that the earth attenuates and changesthe magnetic fields less than the electric fields. The antenna is deployed either on the surface30or underground to communicate with miners under the ground [38]. Broader coverage isachievable by increasing the size of the antenna. Power for the underground transmitter islimited due to permissibility requirements. In this type of communications frequency,geology, noise and depth will influence the probability of successfully communicating withthe surface using TTE communications[38].Figure 2-7 Surface antenna for through-the-earth communications.Based on initial studies, conductivity of the earth?s layers was found to be importantin VLF ranges. Studies showed the presence of any metallic conductor such as pipes, powercables, rails, etc. greatly enhances the transmission of radio waves TTE. Conductivity provedto be a function of the types of rock or soil through which a signal passed as well as afunction of the signal frequency [44]. The effect of rock layers on the EM propagation wasextensively studied by J. Wait et al. [54].Their work underlies later studies on TTE EMprobing or signaling. They have addressed continuous-wave and transient problems over arange of conductivity and dielectric values.Numerous studies of EM noise were also carried out after it was realized that at VLFor ELF ambient noise is a major problem. The source of the background noise was31determined to come from the interaction of particles of solar origin with the earth?s electricand magnetic fields and from worldwide lightning [45]. References concerned with theproblem of EM propagation through-the-earth and other significant accomplishments byJ. Wait et al. are listed in [54]. Currently, TTE signaling is used for emergencycommunications in underground mines. Portable, person-worn wireless TTE systems existand are often used instead of hard-wired systems to establish contact with miners becausethey offer better resistance to damage from roof falls, fires and explosions.2.4.2 Propagation Through-The-WireFor propagation through-the-wire, frequency bands higher than ELF were used bystretching a longitudinal conductor along the tunnel. Such a conductor could support a quasi-TEM mode spread between the conductor and the tunnel sidewall (Figure 2-8a), referred toas the ?monofilar? mode and characterized by a zero cutoff frequency. The fields of such amode are accessible in the whole cross-section of the tunnel at the expense of power loss dueto high power absorption by the tunnel wall. In order to reduce such loss, a two (or more)wire transmission-line (TL) system should be used, whereby a new mode that has anti-phasedcurrents in the two wires is created (Figure 2-8b). This mode, which is usually referred to asthe ?bifilar mode?, has fields that are concentrated in the vicinity of the TL and hence haslower cross-sectional coverage but relatively low loss [55].Under some simplifying assumptions, a modal equation for the monofilar mode of asingle wire in a rectangular tunnel was obtained by Wait et al. [56]. They extended theiranalysis in [56] to derive the modal equations for the monofilar and bifilar modes of a twoopen wire TL inside the rectangular tunnel and found the attenuation constants of thesemodes in a wide range of frequencies. They also considered the excitation of monofilar and32bifilar modes in a TL in a circular tunnel by a short dipole antenna. Based on their results, themonofilar mode was excited more strongly by an antenna placed in the tunnel, but the bifilarmode showed lower attenuation. The excessive losses in the monofilar or coaxial mode areattributed to the return current flow along the tunnel walls.Figure 2-8 a) Current distribution and electric field lines of the monofilar mode (between each wireand the tunnel wall) and b) monofilar mode and bifilar mode (between the wires) [30].In the bifilar or TL mode the fields are more confined to the region between the wireconductors [55]. Monofilar and bifilar techniques ultimately led to the radiating cables andleaky coaxial feeders which have been widely used for underground mine communicationssince the 1970?s. Leaky feeder systems can be obtained by introducing periodicdiscontinuities into the coaxial cable which convert radio frequency energy from a non-radiating bifilar mode to a monofilar mode. The discontinuities are created by the insertion ofspecially designed mode converters, or radiating devices in the cable at the desired intervals[57].332.4.3 Propagation Through-The-AirIn this section, various analytical and numerical models used to characterize TTApropagation in mine tunnels are discussed. As it will be seen, developing analytical modelsfor extreme environments such as underground mines can be very elaborate unlesssimplifying approximations regarding the tunnel geometry are made. These approximationshave been modified over the years according to applications at higher frequencies, andavailability of faster processors. For example, in several models underground tunnels weretreated as tunnels with smooth walls. However, as technology has migrated toward higherfrequencies, analytical modeling has become more sophisticated. As an example, a single-mode waveguide model [11] proposed about forty years ago could model the propagationloss of lower-UHF band signals in mine tunnels. Today?s version has been modified andenhanced to multimode model which is capable of more precisely modeling propagation lossand delay spread in the upper-UHF band. Modeling Tunnels as Hollow Dielectric WaveguidesIn the UHF-band, a tunnel structure may guide the EM wave through the tunnel, andtherefore can be modeled as a waveguide. Inside the waveguides, EM fields can be resolvedinto the sum of propagation modes given by the solutions of Maxwell?s equations subject tothe boundary conditions. These solutions include a dominant mode of propagation with thelowest loss and higher order modes with higher loss. Higher order modes travel at largerreflection angles relative to the waveguide axis (Figure 2-9), and therefore experience morereflections per unit distance and higher losses. Propagation modes of a hollow waveguide, incase of perfectly conducting walls are pure transverse-electric (TE, i.e., Ez=0, Hz?0) andtransverse-magnetic (TM, i.e., Ez?0, Hz=0) modes. Different modes of TEm,n or TMm,n may34propagate in the waveguide depending on the frequency and cross-sectional dimension of thewaveguide.In case of dielectric walls, propagating waves may be represented by hybrid modes ofindex mn, with all three Cartesian components of the electric and magnetic field present [58].These modes are lossy modes because any portion of the wave that radiated on a tunnel wallis partially refracted into the surrounding dielectric and partially reflected back into thewaveguide. The refracted part propagates away from the waveguide and represents a powerloss. By knowing tunnel dimensions and material, Maxwell?s equations subject to boundaryconditions created by the interfaces between the interior of the tunnels and the wall materials,determine the cutoff frequency, propagation constant and propagation loss for each mode.These are important environmental parameters for wireless designs in tunnels andunderground mines.Figure 2-9 Comparison of reflection angles of lower and higher order modes in a waveguide.Early theoretical work on hollow dielectric waveguides with circular and parallel-plate geometries in a medium of uniform dielectric constant was established by authors suchas Marcatili and Schmeltzer [58] and Glaser [59]. Their work became the fundamental basisfor later waveguide-based modeling of tunnels and underground mines [11]. Emslie et al.extended the previous work on waveguides to tunnels by treating them as oversized lossydielectric waveguides with rectangular cross-sections and found approximate mode equations35based on the simple assumption of uniform dielectric constant for the tunnel [11]. Mahmoudand Wait in [11] and Emslie et al. in [11] assumed dielectric constant of the sidewalls wasdifferent from that on top and bottom walls. This provides more accuracy than the simpleassumption of same dielectric constant for all the walls. In [11], Emslie et al. appliedwaveguide model to tunnels with approximately rectangular cross-section, such as coal mineswith considerable degree of roughness, and tunnels with tilted walls. Figure 2-10 shows amap and digital photograph of an underground mine gallery [60].Figure 2-10 Map and digital photograph of an underground gallery [60].They formulated the overall loss for the dominant mode of (m=1, n=1). Overall lossconsists of refraction loss (proportional to f -2), roughness loss (proportional to f -1), sidewalls?tilt loss (proportional to f ), and antenna insertion loss or equivalently antenna coupling lossto the dominant mode (proportional to f -2). Antenna coupling loss occurs due to inefficientcoupling of dipole antennas to the waveguide mode and decreases rapidly with increasingwavelength [11]. At frequencies of interest (UHF), ohmic loss due to the small conductivityof the surrounding material is found to be negligible compared to loss from refractionthrough the walls [11]. Refraction loss has been calculated for both horizontal and verticalCANMETScale 1:100036polarizations of electric field, Eh, Ev, respectively. Depending on whether width or height ofthe tunnel is larger, one of horizontal or vertical polarizations, respectively dominantlypropagate and accordingly only loss of the dominant polarization can be considered.Different losses assuming half-wavelength dipole antennas at both sides in a straight tunnelare given as follows [11]:? ? ? ?? ? ? ?(9)dimensionsal transvers tunnelforcoscos5233.0(8)343.4(7)11343.4(6)1)n(mmodeEfor111343.4(5)1)n(mmodeEfor111343.4020222224422vertical(1,1)232132horizontal(1,1)231312??????????????? ???????????????????????????????????????????????? hywxwhLdLdhwrLdhwLdhwLlossinsertiondipoletiltroughnessrrrrefractionrrrrefractionwhere ?, w, h, ?r1, ?r2, r, d, ?, x and y are wavelength, tunnel width, tunnel height, relativepermittivity of the rectangular tunnel sidewalls, relative permittivity of the rectangular tunnelfloor and ceiling, root-mean-square (RMS) roughness, distance, sidewalls? RMS tilt angle (inradians) about a vertical axis and transversal dimensions (assuming the origin of therectangular coordinate system is on the middle point of the tunnel cross-section),respectively. As it can be seen from the formulas, some losses increase with frequency, andothers decrease, and therefore an optimum frequency can be found in the range 500-1000MHz (Figure 2-11) for minimum overall loss depending on the desired Tx-Rx distance.37Figure 2-11 Overall loss for various distances along a straight tunnel for half-wave antennas withhorizontal polarizations (z is Tx-Rx distance) [11].Tables 2-1 and 2-2 present losses for different Tx-Rx distances and at several UHFfrequencies for a straight tunnel, and a tunnel with a corner, respectively [11]. Lpropagation istotal loss from refraction, roughness and tilt of the walls, and also Linsertion is the half-wavedipole coupling loss which is noticeably high. In case of using antenna with high directivity,coupling loss (insertion loss) which considerably contributes in the overall loss will bereduced. Overall loss is the summation of propagation loss and insertion loss. As it can beseen in Table 2-2, a corner adds an extra loss directly proportional to the frequency. If thetransmitter is outside the tunnel, additional loss due to EM coupling from outside to insideshould also be considered which is not shown in these tables [61]. This loss is dependent onthe distance of the transmitter to the mouth of the tunnel, angle-of-arrival of the wave intotunnel relative to the tunnel axis, cross-sectional dimension of the tunnel and operationfrequency. It should be noted that the results discussed in this part and rest of this chapter arevalid under the assumption of omni-directional antennas at transmitter (Tx) and receiver38(Rx). When using antennas with high directivity propagation is more similar to free spacepropagation rather than waveguide and is predicted to be less sensitive to tunnel?sdimensions and frequency because waves have fewer interactions with tunnels? walls.Table 2-1 Overall loss along a straight path (Eh mode with half-wave antennas) [11].f Lrefraction Lroughness Ltilt Lpropagation Linsertion Loverall (dB)MHz dB/30m dB/30m dB/30m dB/30m dB 30m 150m 300m 460m 600m4,000 0.06 0.05 5.33 5.44 69.90 75 97 124 152 1793,000 0.10 0.07 3.99 4.16 64.88 69 86 107 127 1482,000 0.23 0.10 2.66 2.99 57.86 61 73 88 103 1181,000 0.91 0.21 1.33 2.45 45.82 48 58 70 81 93415 5.34 0.50 0.55 6.39 30.48 37 62 94 126 158200 23.00 1.04 0.27 24.31 17.80 42 139 261 383 504100 92.00 2.08 0.14 94.20 5.80 100 477 948 1419 1890Table 2-2 Overall loss along a path including one corner (Eh mode with half-wave antennas) [11].f Eh Loss per corner Overall Loss (dB) at different Tx-Rx distancesMHz dB 150m 300m 460m 600m4,000 80.2 177 205 232 2593,000 77.6 163 184 205 2262,000 74.1 147 162 177 1921,000 67.6 126 138 148 161415 57.7 120 152 184 216200 47.3 187 308 430 551For the sake of simplicity, Emslie and other authors neglected the continuity of theboundaries of the corner regions by considering different materials for ceiling/floor, andsidewalls [11]. One of the advantages of such approach is that the modeling of tunnels whosewalls have different electrical characteristics is viable [34]. In this model, underground mineswere assumed as rectangular waveguides with perfect geometrical shape, but with lossydielectric characterization. Although this leads to separable Helmholtz wave equation in39Cartesian coordinates, the boundary conditions on the walls necessitate the intrinsicallycoupling of the basic modes and hence propagation constants are not easy to obtain. Asshown by Wait [62], this causes fundamental difficulty in finding the modal eigenvalues andeigenfunctions of rectangular waveguide or any other form than circular. While mostprevious work on modeling tunnels use rectangular waveguides models, circular waveguidemodels have been considered for modeling arched road tunnels [63]-[67].The single-mode waveguide model by Emslie et al. became the basis for laterwaveguide modeling of tunnels. Over time, several researchers tried to enhance the modeland make it more accurate. Because this model only considers the dominant mode to predictthe propagation loss at lower-UHF frequencies, it is only valid for the far-field region wherehigher order modes are greatly attenuated and only the dominant mode exists. Therefore, itfails to predict propagation loss in the near-field accurately. As such, for tunnel microcelldesigns Zhang et al. modified tilt and roughness loss formula of Emslie?s model so that itbecame applicable to near-field [68].The far-field (far-zone) and near-field (near-zone) inside straight tunnels areseparated by a breakpoint [69]. The breakpoint will be explained in more details in the nextsubsection. Near-field waveguide propagation has not been well established but suffers largerloss than far-field propagation. This is because the higher order modes are significant in thenear-field and should be included in calculations.After the breakpoint (in far-field), higher order modes are greatly attenuated andbecome negligible, while the dominant mode remains significant. Far-field waveguidepropagation is stabilized and undergoes smaller loss than near-field propagation [68]. Theattenuation rate of the field in the far-zone is linear in dB, with a slope determined by the40attenuation constant of the lowest order mode (dominant mode) [13]. At higher frequencies(above- UHF) that are much higher than the cutoff frequency of the tunnels, breakpoint isextended further from the transmitter; in other words, the near-field is prolonged. Therefore,to achieve a model that characterizes the near-zone as well, higher order modes should alsobe included [13],[70]. For example, in [13] more than 20 waveguide modes have beenconsidered for accurate modeling at 1 GHz in a railroad tunnel. In [70], Sun et al. proposedthe multimode-waveguide model, which is capable of accurately characterizing fastfluctuations of the channel, and gives analytical expression for the received power and thepower-delay-profile (PDP) at any locations in a tunnel. Based on this model, authors studiedthe effects of various factors, such as size of the tunnel, frequency of operation, electricalproperties of the walls, antenna position and polarization. Two-Slope Pathloss ModelBased on the single-modewaveguide model, pathloss (in dB) increases nearly linearlywith increasing distance in a mine tunnel. However, as shown in Figure 2-12, experimentaland theoretical studies confirm that pathloss has two distinct sections that can be separated bya breakpoint for straight tunnels at UHF-band frequencies. Not having taken this breakpointinto consideration, the waveguide model has overestimated the coverage distance [71].Before this breakpoint, the pathloss shows free space behavior (with free space pathlossexponent,i.e., n = 2) and after it shows waveguide behavior (with lower pathloss exponentthan free space, i.e., n ? 2) [71]-[73]. Accordingly, in [69],[71], Zhang proposed a ray-opticalbased hybrid model for tunnels and mines. This model consists of two types of propagation:(1) free-space model for the region close to the transmitter and (2) waveguide model in theregion far from the transmitter. This was experimentally validated in [71],[74]. The location41of the breakpoint can be obtained by intersecting two pathloss models as suggested in [69].Breakpoint location depends on the tunnel excitation conditions (transmitter inside or outsidethe tunnel). In the case of an external base station, it depends on the angular position of theantenna with respect to the tunnel axis [72]. If the base station is inside the tunnel, thebreakpoint location depends on the antenna radiation pattern, signal wavelength and size oftunnel cross-section. Assuming an omni-directional antenna such as a dipole at bothtransmitter and receiver inside the tunnel rbp, the breakpoint location (or critical distance)mainly depends on the tunnel cross-sectional dimensions (w, h) and the wavelength (?) [73]:2 2 ,bpw hr max ? ?? ?? ? ?? ?(10)Figure 2-12 Measured data for two polarizations at 900 MHz together with the two-slope regression fitsin a coal mine [74].It should be noted that this simple model is only for pathloss modeling and cannotcharacterize the small-scale signal fluctuations of the multimode channel in particular in thenear-field zone. Moreover, this model may not be applied to some short mine tunnels at highfrequencies. For example, in CANMET mine at a 40 m depth for two frequencies of422.45 GHz and 18 GHz, the value of (w2 /?) is 200 m and 1800 m, respectively. Therefore, thetunnel is too short to form a breakpoint and hence no waveguide effect is likely present [75].Consequently, multimode modeling may be considered for propagation modeling of higherfrequencies in short mine tunnels. Ray-Optical ModelsOngoing interest in higher frequency (smaller wavelength) applications has motivatedresearchers in this field to use ray-optical theory for their modeling. Ray-optical models areaccurate when the environment dimensions are much larger than the wavelength and this is acondition that is satisfied in underground tunnels at UHF-band frequencies. In these models,EM waves are considered as optical rays, and EM fields are calculated by summing thereflected rays from the tunnel walls. Unlike modal analysis in the waveguide model, which isrestricted to simple geometries, ray-optical methods can be applied to more complicatedscenarios such as occupied tunnels, tunnels with curvature, coupling between outside andinside of the tunnel, etc.Ray-optical methods based on the classical geometricaloptics (GO) only take intoaccount reflections and not diffractions. Those based on geometrical-theory-of-diffraction(GTD) [76] include both reflections and diffractions, however, the predicted fields at shadowboundaries become infinite, which is impossible in nature and produces a non-uniformsolution. On the other hand, in uniform-theory-of-diffraction (UTD) [77], an extension ofGTD, diffracted fields remain bounded across the shadow boundaries because of the additionof a transition function into the diffraction coefficient.43In [78], Mahmoud and Wait proposed a GO model for rectangular mine tunnels andcompared it with their previous waveguide model for a case of an idealized waveguide withtwo perfectly reflecting sidewalls. Two models showed a satisfactory agreement. This was atheoretical foundation for further analysis. Thereafter, they included the influence of the wallroughness in their model by using theoretical and experimental results obtained byBeckmann and Spizzichino [79] and Beard [80], respectively. In this simple model, the roughsurface is assumed to have a Gaussian distribution and modified Fresnel reflectioncoefficients are considered for the rough surface. The classical Fresnel reflection coefficientis used for smooth surfaces. In [81], a ray-optical model based on GO was developed toinclude curved boundaries. Contrary to classical ray-optical methods where one rayrepresenting a local plane wave front is searched and can only treat reflections at the plane ofboundaries, this ray-optical method is based on ray-density normalization and requiresmultiple representatives of each physical EM wave at a time. Therefore, curved boundariescan also be treated.In [82], authors applied the UTD method to accurately model the coupling betweeninside and outside of the tunnel. This is a critical issue for short road tunnels, where thetransmitter is outside the tunnel and the mobile station is inside with no repeaters betweenthem. This is a case that cannot be easily modeled with waveguide or simple GO basedmodel. A model based on the UTD was also proposed in [83] that allows one to find theeffect of tunnel branches and obstructions such as vehicles and trains in a tunnel. Full-Wave ModelsFull-wave models may also be considered as an alternative method capable of solvingMaxwell?s equations with arbitrary boundary conditions using numerical methods, such as44finite-difference-time-domain (FDTD). The FDTD method is an accurate model that fullyaccounts for the effects of reflection, refraction and diffraction, and provides a completesolution for the signal coverage information throughout a defined problem space. Therefore,it is well suited for accurate study of the EM propagation in complex environments.However, the FDTD requires memory to store the basic unit elements of the model and alsodemands iterations in time in order to update the fields along the propagation direction.Given the large size of tunnels and the high operating frequency (above-UHF), thecomputational burden of conventional FDTD exceeds well beyond the capacity of existingcomputers. Consequently, it has recently been attempted to enhance the efficiency of thismethod for wireless applications in tunnels by employing different approaches to reduce theruntime or computational cost.In [84], excessive computing times were shown to be alleviated via the compute-unified-device-architecture (CUDA) parallel programming route. Whereas in [85], a cost-effective FDTD method for modeling tunnels with realistic construction profiles is proposed,which is based on the compact-FDTD concept and requires minimal computing resources. In[86], the authors have proposed the modified 2D FDTD method that converts a 3D tunnelmodel into a realistic 2D FDTD simulation. This removes the computational burden while atthe same time preserving the factors that form the wireless propagation characteristics. Thismethod has been used to determine a pathloss model that enables effective wirelesssensor-network (WSN) planning and deployment for monitoring and assessing deformation incurved arched-shaped tunnels for Tx-Rx distances of up to several hundred metres. For suchapplications, the FDTD method facilitates accurate modelling of near distance pathloss andclose-to-wall antenna deployments. For most wireless propagation models of tunnels, the45antennas are assumed to be along the central axis of the tunnel. Because this is notrepresentative of most WSN applications where the wireless sensor nodes are mounted on thewalls, it is important to accurately capture the performance degradation resulting from theantenna position using accurate full-wave models. Stochastic and Numerical ModelsIn recent studies on modeling EM propagation in underground mines, more details ofthe environment such as wall roughness are included in order to improve the accuracy. Inmost theoretical models concerning roughness, statistical solutions based on the Gaussiandistribution for random roughness are employed. In [87], stochastic scattering approach ispresented to treat rough surface scattering based on a combination of ray-optical andKirchhoff formulations. Similar to the Kirchhoff modeling, this method is based on atangential plane approximation of the rough surface, i.e., it is applicable to surfaces withgentle undulation whose horizontal dimensions are large compared to the wavelength ofincident waves. However, in contrast to Kirchhoff methods that are only valid for eitherslightly rough or very rough surfaces, this approach simultaneously includes both.In this method as shown in Figure 2-13, each local plane wave front is represented bymultiple discrete rays instead of one ray, in order to model wall roughness. All of thesediscrete rays are reflected back from randomly oriented planes (Figure 2-13b) and not fromthe same boundary plane (Figure 2-13a). In this model, random roughness is characterized bystandard deviation of surface height and correlation length, assuming they follow a Gaussiandistribution. By applying this stochastic scattering approach, the inclusion of random surfacescattering into ray-optical modeling becomes possible.46In [88], a numerical analysis has been used to accurately model roughness andbending in underground mines. In this analysis, the cascade-impedance-method (CIM) andsegmental-statistical-method (SSM) are combined. The CIM method assumes the miningtunnel is a transmission line with diffracting and rough walls (Figure 2-14a). Therefore, itsbehaviour can be considered analogous to a cascade of dielectric impedances with itsassociated losses (Figure 2-14b). In Figure 2-14a, Z?s are the dielectric impedances of thetunnel sidewalls, ceiling and floor. CIM is combined with SSM by dividing the mining tunnelinto segments, each segment into sections and each section into multiple cells in thetransversal and vertical directions. Variation distribution of the rough surface of eachsegment is then simulated by a 3D Gaussian function. From the dielectric impedances of therough walls, equivalent reflection and transmission coefficients of each section in the form ofmatrix can be obtained. This allows the electric field, magnetic field, cutoff frequency andpropagation constant to be determined for each segment of the tunnel.The limitation of this method is in treating the borders of the grid. If the sections arechosen to be infinitely small, the problems of memory space and runtime arise. Therefore, tosimplify the method and overcome these problems, the authors chose to substitute parametersof each section with their average values. As stated by the authors, these types of methods arepreferred over modal analysis (waveguide models) for tunnel mines with rough sidewalls.However, for the case of smooth sidewalls, simple modal theory would be more effective.47Figure 2-13 Stochastic scattering approach: reflection at randomly oriented tangential planes for eachdiscrete ray (a) same plane for all the rays and (b) randomly oriented planes [87].Figure 2-14 (a) A rectangular waveguide and (b) CIM model for the waveguide in Figure (a) [88].48Comparison of Analytical ModelsAs it was seen in this section, different theoretical models have been developed forcharacterizing propagation in underground mine tunnels. In Table 2-3, the main advantagesand disadvantages of each are presented which helps to compare them based on severalcriteria such as complexity, range of validity and modeling capabilities.Table 2-3 Comparison of different analytical modelsSingle-modewaveguide modelMultimodewaveguide modelTwo-slopemodelRay-opticalmodelsStochastic &Numerical modelsAdvantages(1) Simple(2) Providesphysical insight(3) Basis for mostof theoreticalmodelings(1) Accurate fordifferent frequencies(2) Accurate for near-zone as well as far-zone(3) Provides physicalinsight(4) Capable to predictchannel parameterssuch as. RMS delayspread(1) Simple(2) Same accuracy fornear-zone and far-zone(3) Provides physicalinsight(1) Simple(2) Capable ofmodelingtunnel branches(3) Providesphysical insight(1) Similar fordifferentfrequencies(2) Same accuracyfor near-zone andfar-zone(3) Capable ofmodeling tunnelroughness andbranchesDisadvantages(1) Less-accuratefor higherfrequencies(2) Only forpathlosspredictions(3) Only valid forfar-zone of thetunnel(4) Incapable tomodel tunnelroughness andbranches(1) Complex(2) Incapable tomodel tunnelroughness andbranches(1) Fails to modelcases in whichbreakpoint does notexist (e.g., when minetunnel is too short toform breakpoint)(2) Incapable ofmodeling tunnelroughness andbranches(1) Less-accurate forlowerfrequencies(2)Computationalload increasesdramatically asthe signal pathis prolonged(1) Complex(2) Provides nophysical insight(3)Computationallyextensive49As shown in Table 2-3, theoretical models provide valuable physical insights aboutthe EM propagation. However, because most of them are based on non-realistic assumptions,they need to be evaluated experimentally. Although the expense and level of effort forconducting RF measurements increases in complicated environments, the measurement basedapproach has proven to be useful and productive. It will likely remain the principal methodfor characterizing wireless channels in most environments for many years to come. As aresult of its relevance, the next section is devoted to experimental modeling for undergroundenvironments.2.5 Measurement-Based ModelingWhile theoretical models offer physical insights, empirical models are widely used tocharacterize wireless channels because they: (1) are more realistic, (2) provide results that areof immediate use to designers and developers, and (3) are useful in the validation of resultsobtained from simulation-based and theoretical methods. Despite the mentioned advantages,due to the difficulties of access to underground mines, safety issues, measurementcomplexity and expenses, they are not as common in the literature as theoretical studies.Measurement studies for TTA communications at lower- UHF frequencies began innarrowband form with the motivation of characterizing propagation loss. They have beenevolved over time into wideband and ultra-wideband signals with the motivation ofcharacterizing channel-impulseresponse (CIR) and delay spread. As shown in this section,while narrowband pathloss characterization is used to determine coverage area and transmitpower, wideband measurements capture the effects of multipath components bycharacterizing the CIR.502.5.1 Narrowband MeasurementsNarrowband measurements mostly focus on modeling fading statistics andpropagation loss. In this section, we present key findings on characterizing propagation inunderground mines and compare them with their counterparts in conventional surfaceenvironments as well as long tunnels. Fading StatisticsThe distributions, namely, Rayleigh, Rice, Nakagami, Weibull and Lognormal areamong the most commonly used in wireless communications. Rice and Rayleigh are used formodeling LOS and NLOS small-scale fading, respectively, while Lognormal is used forlarge-scale fading above ground. The fading distribution is Ricean (or Rice) when a dominantstationary (non-fading) signal component such as the LOS path is present. As the dominantsignal becomes weaker, the fading distribution will follow a Rayleigh distribution.Components of small-scale and large-scale fading can be separated by applying differentmethods on the NB measured data. For example, the small-scale fading envelope can beextracted from the measured data by normalizing the received signal to its local mean value.Some experimental studies of straight sections in underground mines have shownthey are similar to surface environments; the large-scale fading follows Lognormaldistributions [89], the small-scale fading follows Ricean distribution for LOS scenarios andRayleigh distribution for NLOS scenarios, regardless of frequency [75],[90]. However, somestudies such as the recent one in [89], have reported smallscale fading to follow Rayleighdistribution for some LOS cases in underground mines. This can be attributed to the richmultipath environment formed by the high density of scatterers in the mine.512.5.1.2 Pathloss ExponentSeveral experimental studies have characterized the pathloss exponent for differentLOS and NLOS scenarios in underground mines and compared the findings with otherenvironments. The study reported in [91] shows that at upper-UHF, pathloss exponents inunderground mines are larger than their counterparts in indoor environments [91]. This canbe explained by the fact that in indoor environments, such as in a corridor or a hallway withsmooth walls and clear of obstacles, the pathloss exponent is lower than that of free spacedue to the constructive contribution of multipath signals. In mines, however, the wallsirregularities and roughness are significant, and hence destructively contribute to the signalpower resulting in a pathloss exponent similar to that of free space (n = 2) [91].In another study, LOS and NLOS scenarios for two frequencies (2.4 GHz and5.8 GHz) have been compared. 2.4 GHz showed a lower pathloss exponent than 5.8 GHz forthe LOS scenarios, while 5.8 GHz showed a lower pathloss exponent than 2.4 GHz for theNLOS scenarios [75]. This result shows the difference between propagation in mines and inlong straight tunnels (e.g., transportation tunnels) where increase in frequency decreases thepathloss exponent [92]. This confirms that there are substantial differences betweenunderground mines and transportation tunnels, and therefore may not be treated under samecategory for accurate modeling. Table 2-4 compares and contrasts UHF-band propagationbased on theoretical and experimental studies, in underground mines and two similarenvironments; (1) long straight tunnels and (2) conventional indoor environments.52Table 2-4 Some similarities and differences between underground mines, straight long tunnels andconventional indoor for propagation at UHF-band (f: frequency, n: pathloss exponent, ?rms:RMS delay spread, dTx-Rx: Tx and Rx distance, ?: increase, ?: decrease)Straight long tunnel[70],[92]Conventional indoor[75],[88]Underground mine(experimentalstudy)[75]SimilaritiesLarge-scale fading: LognormalSmall-scale fading: Rice, Rayleighf ?? ?rms (near- region) ?Scatterers and obstacles increase ?rmsLarge-scale fading: LognormalSmall-scale fading for LOS: Rice(Rayleigh has been also reported for somestudies in mines)Small-scale fading for NLOS: RayleighScatterers and obstacles increase ?rmsDifferencesIn underground mines:f ?? n (near-region) ?f ?? ?rms(far- region) ?No correlation between dTx-Rx and ?rmsIn underground mines:No impulse response path arrival clusteringeffectRelatively larger pathloss exponentNo correlation between dTx-Rx and ?rms2.5.2 Wideband/UWB MeasurementsAssuming the wireless channel acts as a linear filter, wideband measurements help incharacterizing it accurately in the time and frequency domain by determining the CIR[42],[93],[94]. From the CIR, the PDP can be obtained, which determines RMS delay spread,received power, time-of-arrival (TOA) of the first path, etc. Providing accurate temporalinformation, the CIR can also be used for location finding applications in undergroundmines. Delay SpreadIn low data rate wireless systems (i.e., when the symbol rate is lower than thecoherence bandwidth of the wireless channel), delay spread can be neglected, and hence53Gaussian noise is the dominant factor that causes bit errors. In high data rate systems on theother hand, delay spread, which causes inter-symbol-interference (ISI), is the main reason forbit errors [42]. As a result, it is necessary to model degrading effects of multipath delay aswell as fading in modern wireless communication systems with high data rates. Bycharacterizing RMS delay spread, coherence bandwidth that is necessary for optimization ofmodulation schemes and data rate, can be determined. Accordingly, wideband measurementshave been conducted in underground mines and tunnels to characterize the CIR. Similar tosurface measurements, many factors such as frequency, wave polarization, height andtransversal (i.e., cross-sectional or bi-dimensional) locations of the transmitter and receiver,scatterers and obstacles between transmitter and receiver affect the CIR [71],[75].For long straight tunnels, the RMS delay spread is found to be: (1) a function of Tx-Rx distance, (2) larger for horizontally polarized waves in tunnels with horizontal aspect-ratio (i.e., width is larger than height) and (3) larger for occupied tunnels [70],[92]. However,some of these results have not been achieved for mine tunnels. For example, no correlationbetween the Tx-Rx distance and the RMS delay spread has been found in mines [75], but hasbeen found to be a function of Tx-Rx distance in both tunnels and indoor environments. Thefunction is an increasing function at first and a decreasing function after a certain point (dual-slope relation) [70]. Some of these similarities and differences between mines and tunnelsand conventional indoor environments are listed in Table 2-4.RMS delay spread in mines has been found to be highly dependent on bidimensionalposition of the antennas [60] and larger in a mine with more rough walls, branches andobstacles [60],[92]. It has also been compared for different frequencies. In [75], two WLANfrequencies of 2.4 and 5.8 GHz have been compared. This study found the RMS delay spread54to be larger for 2.4 GHz and they concluded that the maximum usable data rate with arelatively simple transceiver would be higher at 5.8 GHz in the mine they conducted theirmeasurements.In Table 2-5, typical values for the pathloss exponent and RMS delay spread ofdifferent environments are presented. This allows the reader to compare underground mineswith other environments more conveniently. It can be seen that for underground mines, thepathloss exponent and RMS delay spread have been found in the range of 1.8-5.49 and 1.7-60 nsec, respectively. The range of values varies according to the measurement scenario, sizeof the tunnel and frequency. A more detailed comparison of experimentally determinedUWB and fading statistics experimental between mines and different indoor scenarios can befound in [89]. Location FindingCIR characterization achieved through WB measurement can also be used toaccurately (to within 2 m) locate mobile stations in mines and other confined environmentsfor location tracking applications. As an example, the CIR provides the required informationfor using the fingerprinting technique, which in conjunction with a neural network canaccurately locate the mobile [95]. Each user?s information, such as CIR and PDP, is afunction of user?s location and can be obtained by several offline wideband measurements.These are recorded in a user fingerprint database and subsequently compared to real-timemeasured fingerprints corresponding to the user?s new location [95].55Table 2-5 Pathloss exponent and delay spread assuming omni-directional antennas for severalfrequencies in different environments.Type of EnvironmentMeasurementfrequency(GHz)Size(width ? height ?length)Mean ?ms(RMS-delayspread)in nsecPathlossexponent(n)ReferenceFree Space ------- ------- ------- 2[42],[93]Urban AreaCellular RadioUHF-bandHundred metres(an outdoor site)-Several kilometers(San Francisco)40-255002.7-3.5Shadowed Urban AreaCellular Radio3-5In-Building (LOS) Typical office-Open planfactory4-1301.6-1.8Obstructed In-Building 4-6SeveralStraightLongTunnelsEmpty0.9Tunnel1:  3.34?2.6?259Tunnel2:  7.5?4?2000Tunnel3:  4.2?3?10,000Tunnel4:  2.4?2?200Tunnel5:  7?3.7?120Less than 251.87-5.49[92]Occupied Less than 103StraightLong TunnelEmpty0.9 3.34?2.6?2594.124.2-4.49Occupied 21.7StraightLong TunnelEmpty1.8 3.34?2.6?2596.032.12-2.46Occupied 58.65UndergroundMine(Level 70m)LOS (1-10m)UWB  (3-10) (2.5-3)?3?701.72 1.8[94]NLOS (1-10m) 3.76 4.01Underground Mine(Level 40) NLOS & LOS2.4 5?5?75 Less than 60 2.13-2.33 [60]UndergroundMine(Level 70)LOS (d=1m)2.4(2.5-3)?3?706.342.03[75]NLOS(d=23m)4.62UndergroundMine(Level 70)LOS (d=1m)5.8 5.112.22NLOS(d=23m)3.51UndergroundMineLOS (d=40-44m) 0.4-0.5 5?6?50019 ------- [12]NLOS 25-42 ------- [12]56The database, however, should be updated due to the fact that mines are dynamicenvironments [50],[96]. Heavy machinery or moving objects may considerably change theproperties of the channel, requiring an update of the database?s information (e.g., a newtraining of the neural network). This channel variation issue can be resolved by using amaster neural network [50], [96].2.5.3 Multiple-Antenna MeasurementsIn addition to experimental studies of single-antenna systems in underground tunnels,multiple-antenna systems such as multiple-input-multiple-output (MIMO) systems have alsobeen studied. Because there have been no MIMO studies in mines to date, results presentedhere are from experimental studies in transportation tunnels that are relatively similar to minetunnels. MIMO measurements for underground tunnels were originally motivated by interestin supplying GSM-Rail service at 900 MHz [21], and most recently for advanced WLAN,worldwide-interoperability-for-microwaveaccess (WiMAX) and long-term-evolution (LTE)service at 2 GHz, in transportation tunnels. It is a well-known fact that performance of aMIMO system is mostly affected by the correlation between fading observed on adjacentantenna elements. This correlation depends on the type and configuration of the antennas andthe range of angles over which the signals arrive or depart (quantified by angular spread oftransmitter and receiver).Despite the small angular spread of the direction-of-arrival (DOA) and direction-of-departure (DOD) of the rays in the tunnels, preliminary experimental results have shown thatmultiplexing gain (or capacity gain) is achievable by employing multi-antenna techniques[21],[24],[97]. However, the channel capacity is strongly dependent on the Tx-Rx distance,tunnel size and geometry. In a MIMO study in [24], correlation distance of antenna elements57is found to be an increasing function of the transmitter-receiver separation, and is larger forreceiver elements in tunnels of smaller sizes. Correlation distance is the average spacingbetween two neighboring antenna elements at one end that produces a correlation coefficientsmaller or equal to a certain value, typically 0.7. Antenna spacing in a MIMO system shouldnot be less than the correlation distance to achieve acceptable performance. Despite thevaluable contributions of MIMO studies in underground tunnels, MIMO performance is stilluncertain in underground mines because of physical differences. This gap in the research,along with the promising results from MIMO in transportation tunnel studies, may encourageresearchers to consider MIMO in underground mines for their research.2.5.4 Techniques to Overcome Channel ImpairmentsIn this section, we will discuss required baseband modeling considerations, whichshould be taken into account while designing underground mine radio receivers. In thisregard, the main focus is on combating key channel impairments such as multipath fading(causing fading), delay spread (causing ISI) and Doppler spectrum (causing inter-carrier-interference: ICI). Experimental characterization of channel impairments can be very usefulin radio receiver designs.To eliminate ISI in underground environments, the orthogonal-frequency-division-multiplexing (OFDM) technique that is well-known for high data-rate wireless transmissionsand robustness to multipath delays is used [98],[99]. OFDM is intrinsically capable ofcombating common distortions in the wireless channels without requiring complex receiveralgorithms. Compared to conventional single carrier techniques, OFDM-based systems havea low complexity implementation in which instead of a complex equalizer, channel58estimation based on the CIR is used to recover the received signal. The vector CIR can berepresented by the following formula [100]:? ?)()()( )(1ttetAth itjKiii ??? ????(11)where i is the number of multipath components, Ai(t) is the amplitude of the ith path, ?i(t) isthe time delay of the ith path and ?i(t) is the phase shift of the ith path. A widebandexperimental characterization in CANMET mine [101] reveals that TOA of paths follow amodified Poisson distribution and their amplitudes undergo Rayleigh or Rice fading withuniformly distributed phase over [0, 2?]. Based on this statistical information and using theabove formula, an OFDM channel estimation method is studied for wireless LANcommunications in underground mines [99], which employs the pilot-symbol-assisted (PSA)method. Performance of different estimation algorithms and modulation schemes such as 16quadrature-amplitude-modulation (16QAM) for a 24 Mbps link, quadrature-phase-shift-keying (QPSK) for a 12 Mbps link, and binary-phase-shift-keying (BPSK) for 6 Mbps link,derived from the IEEE-802.1l wireless-LAN standard, are assessed and compared in terms ofthe bit-error-rate (BER).To combat multipath fading for WSN applications in underground mines, the chirp-spread-spectrum (CSS) method is proposed [83]. The CSS uses wideband linear frequencymodulated chirp pulses to encode information. It is resistant to channel noise, multipathfading even when operating at very low power and Doppler shift for mobile applications. TheCSS method is suitable for wireless personal and sensor network communications, whichrequire low power usage and need relatively low data-rates (1 Mbit/s or less).59Rather than combating, multipath components can also be exploited effectively inunderground mines for increasing SNR based on diversity combining methods. This can beachieved by combining the energy in various multipath components using a rake receiver. ARake receiver is not considered feasible in other industrial environments due to the largenumber of fingers required to combine the many resolved multipath components [102],[103].In underground mines, however, the energy is concentrated in fewer multipath components,and hence the number of fingers required is far less [104]. Study in [99] shows that theOFDM channel estimation that performs well at low Doppler frequency can efficientlyreduce the effect of Doppler shift in underground mines. Nevertheless, due to low vehiclespeeds, i.e., typical in underground mines, system performance is less affected by ICI, andtherefore ICI can be neglected [99]. Unlike underground mines where Doppler spectrum isnegligible, it is one of the key channel impairments for vehicular-to-vehicular (V2V)applications in subway tunnels. This should be considered while designing vehicular wirelesscommunication systems such as intelligenttransportation- systems (ITS). In [105], a V2Vwireless channel inside a large subway tunnel has been experimentally characterized. It isshown that the V2V fading process is inherently non-stationary, and based on theestimations, RMS delay and Doppler spreads are log-normally distributed. Their studyreveals that the spreads, excess delay, and maximum Doppler dispersion are larger onaverage when both vehicles are inside the tunnel compared to the ?open-air? situation.Satisfying the institute-of-electrical-electronics-engineers 802.11p (IEEE-802.11p) standardrequirements, they concluded that this standard will be robust towards inter-symbol-interference (ISI) and ICI inside a tunnel.60As was seen in this section, measurement-based characterization of undergroundmines provides realistic results that are of immediate use to designers and developers, andtherefore has attracted more attention in the last decade. However, before applying theresults, one must ensure that measured data is sufficient for any statistical inference about thepropagation environment. This can particularly be a concern for underground mines becausemine galleries differ from mine to mine, and even from level to level within the same mine[94]. Inconsistent experimental results from different UWB measurements in mines confirmthis fact [94]. As a result, more measurement campaigns are required to achieve more generaland non-site-specific conclusions about propagation in underground mines and theirgalleries.2.6 Implications for Wireless Communication System DesignDesigners and developers use channel models to predict and compare theperformance of wireless communication systems under realistic conditions and to devise andevaluate methods for mitigating the impairments and distortions that degrade wirelesssignals. In this section, general conclusions achieved from characterization of wirelesspropagation in underground environments are presented.We will see how tunnel dimensionand geometry, wall material and obstructions affect parameters such as optimum frequency,attenuation, RMS delay spread and angular information that are required in effective designof wireless systems. Additionally, we will see how antenna properties may be affected byunderground geometry. This information is particularly valuable when configuring designsfor multiple antenna systems.612.6.1 Frequency of OperationThe optimum frequency for TTA communications in underground mines depends onseveral factors, including tunnel size, tunnel geometry and infrastructure inside the tunnel.At the frequencies below UHF-band (MF-VHF), signal propagation is enhanced viacoupling to conductors that may be in the mine entries, and antenna efficiency is notnecessarily compatible with sizes that are portable [31]. At UHF frequencies on the otherhand, theoretical studies show that mine tunnels act as relatively low-loss dielectrics withdielectric constants in the range 5-10, and therefore transmission takes the form of waveguidepropagation in the tunnel [11]. As it has been shown by Emslie et al. in the 500-1000 MHzrange, attenuation is relatively low in straight mine entries [11].In contrast, practical tests have revealed that the MFband (300 kHz-3 MHz) has moredesirable coverage with less severe attenuation compared to UHF-band in both coal andmetal/non-metal mines [38]. The MF-band has a proven coverage area of 300-460 m inconductor-free areas, and as much as 3200 m in conductor-filled areas where parasiticpropagation help the signal travel longer distances [38]. Higher frequencies such as VHF,UHF and SHF propagate in LOS and 300 m down a mine entry. However, it is unlikely thatan unaided (i.e., no leaky feeder) VHF or UHF signal would be able to travel around morethan about two crosscuts [38].This contradiction can be explained by noting that attenuation rate of UHFfrequencies is lower than MF frequencies assuming that the tunnel has smooth walls, isstraight and empty. This is an unlikely situation for most underground mines and as it hasbeen shown that attenuation of UHF frequencies is significantly higher when the signal62propagates around a corner or when obstacles such as massive piece of machinery is in thepath of propagation [31].As a result, considering practical tests, and theoretical and experimental results,although high frequencies (UHF and higher) may offer a larger coverage area in straight andunobstructed tunnels, better coverage may be achievable by frequencies lower than UHF(MF-VHF) [30],[38] when corners, crossings and obstacles exist. Regardless, UHFbasedtechnologies are more appealing to the mining industry because low cost, small form factor,scalable and easy to use applications are available off-the-shelf. In addition, their coverageand other propagation issues (e.g., requiring lineof- sight between Tx and Rx) can beresolved by appropriate antenna and wireless network designs.2.6.2 Tunnel GeometryIn this subsection, we discuss how tunnel structure impacts wireless communicationsin underground mines. In addition to pathloss, extra losses in tunnels are introduced due totunnels? curvatures, sidewalls? tilt and changes of the cross-sections. Cross-Sectional DimensionBecause each tunnel behaves like a waveguide, its cross-sectional dimensiondetermines the cutoff frequency. For tunnels of arbitrary shape, a very rough approximationof the cutoff frequency is determined by the frequency whose free-space wavelength is aboutor equal to the tunnel perimeter. Well above this cutoff frequency, the number of propagatingmodes grows by the square of frequency. The cross-section of most tunnels that canaccommodate vehicles, has dimensions of a few metres and the cutoff frequencies areconsequently of a few tens of megahertz [106].63In addition to cutoff frequency, cross-sectional size also impacts the attenuationconstant and small-scale fading statistics, owing to the fact that change in tunnel cross-sectional size is equivalent to change in frequency for TTA propagation. As such, an increasein a tunnel?s width and height increases small-scale fading, which is similar to the impact ofincreasing frequency. For tunnels with larger cross-sectional dimensions compared to thewavelength, multipath becomes more significant, leading to more severe fluctuations andgreater fading [11],[31],[71],[107],[108]. In addition, in larger dimension tunnels theattenuation constant is smaller, and therefore the nearfield zone with small-scale and deepfluctuations is prolonged [70]. Severe small-scale fading for larger size tunnels (or higherfrequencies) implies to either include fading mitigating techniques for SISO systems orconsider MIMO systems to benefit from rich multipaths.Cross-sectional size also impacts EM polarization loss. Propagation losses forhorizontal and vertical polarizations are relatively the same in tunnels with a square cross-section with aspect-ratio of 1 (aspect-ratio: ratio of longer dimension to the shorter one) [92].However, in rectangular tunnels with larger width than height (horizontal aspect-ratio),horizontal polarization has less attenuation in the region far from the transmitter. Curvature, Junctions and TiltAt the UHF-band, curvature in tunnels introduces additional loss. This loss isdependent on frequency of operation, width of the tunnel, radius of curvature (Figure 2-15)and wave polarization. This loss is inversely proportional to the radius of curvature, and incontrast to straight tunnels, it is linearly proportional to frequency and width of the tunnel[109]. In curved tunnels, because the horizontal electric field is perpendicular to the curved64walls, horizontally polarized waves are much more affected by the tunnel curvature thanvertically polarized waves [13].Junctions and bends change wave polarization. Therefore, unlike straight sections,inside the junctions/bends (or around the corners) waves are depolarized, and thereforereceived power around corners is usually independent of the receiver antenna?s orientation[31]. The loss associated with one single corner is given in Table 2-2. As it can be seen,corner loss is smaller for lower frequencies. Additional corners add extra loss and increasethe overall loss. Attenuation caused by corners in mine tunnels can be compensated for byadding 90o reflectors to extend wireless propagation beyond the junctions and corners [110].In most mine tunnels, walls are tilted about a vertical axis (tilt angle), which adds extra loss.Tilt loss becomes more significant as frequency increases [11], and therefore demands moreprecise analytical modeling at higher frequencies [111]. To precisely model pathloss andlarge scale fading, this loss should be included in total loss for tunnels with considerable tilton sidewalls, floor or ceiling.Figure 2-15 Radius of a curvature in a rectangular curved tunnel [13].2.6.3 Surface RoughnessUnderground mine tunnels have wide variations in wall roughness, often on the orderof 20 cm [13]. However, unlike tilt and curvature loss in a straight empty tunnel, roughness65loss becomes more significant at lower frequencies in tunnels owing to the larger grazingangle and higher number of bounces per unit length [11],[111].Therefore, this loss isinsignificant at high frequencies (UHF and above) compared to tilt, curvature, corner andobstacle losses as Table 2-1 indicates.2.6.4 Material and InfrastructureBecause infrastructure inside the tunnel changes EM properties of the tunnel, it istreated in the same section as the material from which the tunnel is dug. MaterialElectromagnetic properties of different materials are characterized by parameterssuch as relative permeability ?r, dielectric constant ?r and conductivity ? [106]. As shown in[11], Emslie et al. formulated the overall loss for UHF frequencies by assuming that theeffect of ohmic loss is negligible due to the low conductivity of the surrounding material(coal) in coal mines. Conductivity of the coal at MF-band frequencies is 3 ? 10?5- 4 ? 10?5(S/m) and at 9 GHz is 0.12-0.73 (S/m) [106]. Dielectric constant of the coal at MF-bandfrequencies is 10-34, and at 9 GHz is 3.4-3.9, respectively [106].Unlike permeability that for most rocks (except for rocks with a high concentration ofmetals of the ferromagnetic group) is very close to that of vacuum ?0 = 4? ? 10?7 H/m, thedielectric constant and conductivity of rocks, are highly variable. Dielectric constants rangefrom 2 to 70, but more frequently from 4 to 10 and conductivities range from 10?6 to 1 S/m.In general, both ? and ?r increase with the water content and, as a general rule, ? increasesand ?r decreases with frequency [106].66In spite of the high variability of dielectric constant and conductivity of rocks, mostrecent studies such as [13],[70] show that walls? material does not significantly impactwireless propagation inside the mines, and as frequency increases (UHF and above), thematerial shows weaker impact. For WSN applications [86] however, material has been foundto be an important element to consider in tunnel wirelss propagation. This can be due to thesmall spacing of sensors and tunnel walls. In practical deployments, wireless sensors areoften attached to walls with an antenna to wall spacing of less than 10 cm. Nevertheless,compared to antenna position and frequency of operation, this study concludes that materialhas less impact on wireless propagation in tunnels. Unlike insignificant impact of material atUHF-band for TTA communications, at medium frequency (MF) for TTW and TTEcommunications, material has significant impact on wireless propagation. In TTEcommunications, earth is the medium for propagation, and in TTW communications, the skindepth in the mine?s wall for the return current of the monofilar mode is the degrading factor.That explains why there are many theoretical and experimental studies on the effect ofmine?s material on early underground communications. InfrastructureConductors such as power cables, water pipes, rails, etc. inside the tunnels,considerably change the electromagnetic properties of the tunnel, in particular for TTE andTTA communication [106]. Metallic air ducts or heating-ventilating- and-air-conditioning(HVAC) systems that are used to circulate air within the underground mining complex mayenhance the wireless propagation. Additionally, ground support infrastructure, such as wiremesh screens, used to prevent rock from falling and the tunnel from caving-in can also affectwireless propagation. At corresponding frequencies, where the mesh netting interval is on the67order of 0.1 free-space wavelengths, the attenuation for the dominant propagation mode canbe reduced [112].2.6.5 Vehicles and Other ObstructionsHeavy machinery, trucks, miners and other obstacles increase propagation loss andRMS delay spread in tunnels. This loss is dependent on the dimension of obstacles. Largerdimension vehicles cause additional shadowing loss [108]. In this case, propagation rays mayonly find their way behind the vehicle through multiple diffractions. This degrades signalpower significantly [83]. In addition to power loss, obstacles increase RMS delay spread, andtherefore a decrease in data rate can be seen in occupied tunnels. As shown in Table 2-3, inan experimental study, RMS delay spread was found to be less than 25 ns for a vacant tunneland 103 ns for the same tunnel when occupied [92]. This demonstrates the influenceobstructions have on delay spread. These results should be taken into account when decidingon transmit power, symbol rate, etc. for wireless transmission inside mines.2.6.6 Antenna Placement, Gain and PolarizationThe pattern and polarization of transmitter and receiver antennas greatly impactswireless propagation in confined environment such as mines. As can be seen in Table 2-1,antenna insertion loss contributes significantly to the overall propagation loss for all of thestudied frequencies. The importance of antenna parameters becomes more evident if multipleantenna systems are being used. In this case, not only the received power but also thedecorrelation (or orthogonality) of antenna elements should be considered. Because there areno MIMO experimental studies in underground mines to date, some of the materialspresented in this section are extracted from experimental studies of multiple-antennas in68underground subway tunnels, which can help in predictions for employing MIMO systems inunderground mines. Antenna PlacementWhile the attenuation rate is mostly determined by the tunnel size and operationfrequency (and not the location of the antennas), the power distribution among propagationmodes is governed by the position of the transmitter antenna [70]. This can be valuableinformation for determining transmitter antenna position for different wireless systems withdifferent objectives. For example, in single-input-single- output (SISO) systems, higherpower can be achieved by positioning the transmitter antenna at the centre of the tunnel,while for MIMO systems, studies have shown that positioning the transmitter off-center ofthe tunnel crosssection would offer higher capacity [24].Cross-sectional location of the antenna also impacts the radiation pattern byintroducing additional loss [92]. Antennas designed for free-space may not perform well inunderground mines because of the large number of strong reflections from the walls, this hasbeen regarded as insertion loss (or coupling loss) that was priorly discussed. This effect ismore noticeable for omni-directional antennas, such as dipoles, when they are located off-centered of the tunnel cross-section. In fact, omni-directional antennas perform better (moresimilar to their behavior in free-space) when they are in the middle of the tunnel.It is also shown in [21] that alignment of the antenna array in MIMO systems plays acritical role. MIMO configurations perpendicular to the tunnel centerline or along a diagonalline, are found to be better than parallel to the tunnel centerline.69Locating the transmitter outside or inside of the tunnel affects the radiation pattern ofthe Tx antenna. When the transmitter antenna is inside the tunnel, its radiation patternbecomes sharper than its free-space radiation pattern (i.e., radiation pattern that an antennawould have if it were in free space where there is no reflection, scattering and diffraction).This change does not depend significantly on the transmitter position along the tunnel.Conversely, when the transmitter is outside, the coupling loss between free-space and insideof the tunnel is dependent on the transmitter position. The coupling loss is considerablydependent on the angle of incidence too, if it is greater than 10o with respect to the tunnelaxis [82]. Antenna Gain and PolarizationWhen an omnidirectional antenna is located inside the tunnel, its effective radiationpattern becomes sharper than its free-space radiation pattern. Results in [110] show that adirectional antenna located in the middle of the cross-section of the tunnel is a desirablechoice for effectively transmitting power along the tunnel. However, locating the antenna inthe middle of the tunnel may not be practical for most cases.For SISO systems in tunnels, signal-to-noise-ratio (SNR) may be improved byemploying a directional antenna and directing the antenna beam parallel to the centerline ofthe tunnel because it prevents signal scattering in LOS scenarios. However, in NLOS caseswhere propagation into the branches is dependent on multipath components, directionalantenna may not be a good option.For MIMO systems in which a rich scattering environment is beneficial (due tooffering larger angular spread), directional antennas may not be a good alternative unless, as70shown in [21] its beam-width is equal to the angular spread. Preliminary measurements in[21] have shown that the average value of the angular spread in a subway tunnel is around60o, which can be different in underground mines due to their smaller size, roughed walls andinfrastructure. For horizontally polarized antennas, the tunnel width plays a more significantrole as the reflection coefficients on the horizontal ceiling/floor are larger than those on thevertical walls. Likewise, the tunnel height is more significant for vertically polarizedantennas [70]. In rectangular tunnels with larger width than height (horizontal aspect-ratio),horizontal polarization has less attenuation in the region far from the transmitter. This can beinterpreted from Emslie?s refraction loss formula for both polarizations, and its experimentalverification in [107]. As a result, horizontally polarized antennas are more appropriate inwide but short tunnels, and vertically polarized antennas are more appropriate for tall tunnels[70]. In addition, for a tunnel with horizontal aspect-ratio (width: 4.2 m, height: 3 m) theRMS delay spread is larger for horizontal polarization (13.5 nsec) compared to verticalpolarization (5.49 nsec) [71]. Different RMS delay spreads are caused by differentattenuation constants for the two polarizations. The lower attenuation of the horizontallypolarized signals makes the power-delay-profile spread [71].As it was seen in this subsection, optimizing antenna performance in tunnels remainsan open topic for future studies. Better performance can be accomplished through carefuldesign of customized antennas that are matched to the tunnel environment. In Table 2-6,significant results and implications useful for wireless system design have been presented.This table demonstrates how operation frequency, physical parameters of tunnels and antennaproperties affect two main channel impairments; power loss and RMS delay spread. (parts71left with ?NA? (not-available) representing cases that have not studied to date, or noinformation was available).Table 2-6 Key implications for wireless system design at UHF-band in underground mines.Loss / Attenuation RMS delay spread ReferenceTunnelGeometryCross-sectional size(similarimpact asfrequency)Larger size at a fixed frequency (orhigher frequency in a fixed cross-sectional size)shows higher pathloss exponent inLOS scenarioLarger size at a fixed frequency (orhigher frequency in a fixed cross-sectional size)shows smaller  RMS delay spread[75]Curvature/corner/branchIncrease(more significant at higherfrequencies)Increase [55],[71]Wall tiltIncrease(more significant at higherfrequencies)NA[11],[55]Wall roughnessIncrease(more significant at lowerfrequencies)IncreaseWall material No significant impact No significant impact [13],[70]ObstructionsPresence of obstruction results inhigher lossPresence of obstruction results in largerRMS delay spread[71]AntennapropertiesCross-sectionalplacementCentre of the tunnel shows lower losscompared to off-centreCentre of the tunnel shows larger RMSdelay spread compared to off-centre[70]Radiationpattern (gain)Directional antenna shows lowercoupling loss compared to omni-directional one in LOSDirectional antenna shows smaller RMSdelay spread compared to omni-directional one[11],[94]PolarizationMine tunnel with horizontal aspect-ratio shows lower loss forhorizontally polarized wavesMine tunnels with horizontal aspect-ratioshow larger RMS delay spread forhorizontally polarized waves[12],[50]2.7 ConclusionsThe need for wireless communication in the underground mining industry hasevolved from basic emergency signaling, to person-to-person voice communication and tohigh speed real-time data transmission. Accordingly, the supporting technologies haveemerged from through-the-earth transmission, to radiating cables, to point-to-point and multi-72point radios. Applications utilizing these technologies include voice communication, videosurveillance, tele-operation of mining equipment (tele-mining), wireless sensors networks,geo-location and tracking of personnel and assets. To develop and evaluate thesetechnologies appropriately, wireless propagation and channel models are essential.Measurement and theoretical approaches to channel modeling are increasingly seen ascomplementary; many channel modeling studies employ both methods.Analytical and numerical models based on waveguide theory, geometrical optical ray-tracing and other methods have been developed by many researchers. While the single-modewaveguide model is simpler and requires fewer inputs about the physical environment, it isnot very effective in predicting propagation for near-field and too short tunnels with complexgeometries at higher frequencies. Ray-optical models on the other hand, provide moredetailed prediction for higher frequencies and complex geometries at the price of requiringdetailed information about the physical environment, and computational burden whichincreases significantly if the area under study is prolonged. A recent theoretical model,multimodewaveguide, offers more accurate and realistic model with reasonable runtime,which can also characterize small-scale fading statistics. The main advantage of this model isthe ability to accurately characterize both the near-zone and farzone of the tunnel.Experimental studies on the other hand, provide readily usable parametric values butare site specific and their statistical generalization requires extensive measurementcampaigns in the underground mines. Measurement results in different frequency bands (e.g.,VHF, UHF and SHF) have been presented by several researchers, mostly coveringnarrowband transmission. Some limited wideband and UWB studies in UHF and SHF bandshave been conducted. With the increasing interest in low-power sensor networks and73batterypowered access points, the availability of UWB transmission technologies and thetransition towards even higher, millimetre wave frequencies for indoor applications, the needfor further theoretical and experimental studies in underground mines is imminent.Some implications of wireless communication design have been discussed in thisarticle, which are interpreted from the previous studies. Open research areas for futureinvestigation include characterization of frequencies above 10 GHz, incorporation of morecomplicated mine geometries in existing waveguide and ray-optical models, antennaconfiguration design and development of channel models for MIMO systems, design ofoptimum wireless mesh networks, channel modelling for body-area-networks, and etc. Theresults of these studies facilitate the employment of new technologies by the mining industrythat ultimately improves work safety, productivity and efficiency in mines.74CHAPTER 3: MIMO EXPERIMENTAL SETUP AND DATACOLLECTION3.1 IntroductionIn this chapter, we describe the channel sounder we have developed and usedthroughout our experimental study. Both the measurement hardware and software that weused to collect the data and the method that we used to calibrate the setup are presented andexplained.  This channel sounder is capable of collecting data for both MIMO and UWBmeasurements.3.2 Development of a MIMO Channel SounderIn this section, we briefly describe the UWB-MIMO channel sounder that wedeveloped for our MIMO measurements in the Radio Science Lab (RSL). It includes both thechannel sounder hardware and the data acquisition software. Because in parallel to ourMIMO study, another project concerning UWB channel modeling in underground mines wasunderway, this channel sounder was developed so that it can collect both UWB and MIMOdata at the same time.3.2.1 MIMO Channel Sounder HardwareMIMO channel sounders generally fall into two categories: those based on real arraysand those based on virtual arrays. In a real-array-based channel sounder, all NRx receivingantennas (and possibly all NTx transmitting antennas) are present and used simultaneously. In75a virtual array-based channel sounder, a single transmitting antenna and a single receivingantenna are moved to each of the locations at which a MIMO antenna element would beinstalled and individual channel responses are measured sequentially. This has severaladvantages: (1) mutual coupling between antenna elements is absent and (2) only one RFchain is required. However, the measurement environment should be static. The condition ofhaving a static environment was met in both the service tunnel and the Myra Fallsunderground mine measurements. Both the service tunnel at UBC and level 23 of the MyraFalls underground mine where we conducted the measurement were vacant, and there was nomovement.For simplicity as well as versatility, we decided to build a virtual array-based MIMOchannel sounder. The basic MIMO measurement setup (without fibre-optic cable) can beused for distances of up to 20 m. The principal limitation is the loss and phase distortionintroduced by overly long coaxial cables. Possible approaches for extending the channelsounding range are using expensive phase-stable coaxial cables, and/or phase-locked remotesignal generators.  However, none of these options is easily scalable for ranges varying fromtens of metres to a few kilometres.We extended the measurement range of our channel sounder to a few hundred metresby incorporating a Miteq RF-over-fibre unit into the transmit side. Compared to coaxialcable, fibre optic cable is also much less susceptible to phase distortion due to cable torsionand flexion and incurs far less path loss. A block diagram of the measurement setup and themeasurement specifications are given in Figure 3-1 and Table 3-1, respectively.76Figure 3-1 Block diagram of the measurement setup.Table 3-1 Measurement specificationsUnit SpecificationVNA master Anritsu MS2034AFrequency bandNumber of frequency points2.49 - 4 GHz551Antenna Electrometrics UWB Biconical antennasPositionerTx powerVelmex linear xy-positioner26 dBmGrid point spacing 2? at 2.49 GHzDynamic rangePower amplifierFibre optics120 dBOphir - Model 5303075Miteq SMCT-100M11GMeasurement setup consists of an Anritsu MS2034A master vector network analyzer,two antennas (Biconical for UWB measurements), coaxial cables, two Velmex bislide xy-positioners with motor controllers, two USB-to-serial adapters, a laptop-based instrument77controller running MATLAB and three carts to carry: (1) Rx positioner and VNA, (2) Txpositioner and Fibre Optic spool and (3) power supply (inverters and batteries).Figure 3-2 MIMO Measurement equipment in RSL.A MATLAB program is used because of its post processing and peripheralcommunication abilities. It supports the general-purpose-interface-bus (GPIB) functionality.GPIB is the interface between the laptop and general purpose-network -analyzer (PNA). It isa widely used industrial interface between instruments. Here, it serves as a link between thelaptop and PNA for sending and reading data or commands. After the GPIB connection isestablished, the command that tells PNA what to do needs to be sent. The standard-commands-for-programmable-instruments (SCPI) command format is used here.78In order to make the channel sounder self-contained and easy to transport, wemounted the equipment together with storage batteries and a true-sine-wave inverter in twoaluminium carts that are equipped with 0.5 m pneumatic tires. Before conducting ourdevelopment run in the mine, we calculated the link budget, calibrated the measurementequipment and performed some tests in a hallway of the MacLeod building at UBC in orderto ensure that the whole system functions correctly.For UWB-MIMO measurements in both the UBC service tunnel and the Myra Fallsunderground mine, we used Biconical antennas which have 3.3 dB gain difference in itspattern for frequencies under the study, 2.49 GHz - 4 GHz (-1.9:1.4 dBi).  This gaindifference does not impact our narrowband MIMO analysis.3.2.2 Calibration of the Channel SounderIn order to cancel the effect of different electrical components such as cables andamplifiers in the circuit, a calibration procedure should be performed. There are differentways to perform the calibration; the most accurate one is full 12-term error correction.Because the RF over fibre link is uni-directional, we cannot perform a full 12-term errorcorrection. Moreover, due to limitations imposed by the VNA firmware, we cannot performsix-term error correction (impedance of port 1 and forward gain). Therefore, we chosefrequency response correction for the calibration, as shown in Figure 3-3.After calibration is finished and saved, cables can be disconnected from the bulletconnector, and can be reconnected back to antennas respectively.79Figure 3-3 Connection diagram for calibration.3.2.3 MIMO Data Acquisition SoftwareAs aforementioned, a MATLAB script that runs on the laptop-based controllercontrols the MIMO channel sounder. During a typical measurement, the transmitting antennais horizontally translated to one of the NTx transmitting grid points and then channelfrequency response data is collected after the receiving antenna is relocated to each of the NRxreceiving grid points. The process is repeated NTx times.The complete measurement procedure as shown in Figure 3-4 can be summarized asfollows: (1) Set the measurement parameters such as start and stop frequencies, number ofsample points, IF frequency and RF transmit power. (2) Initialize the xy-positioners: (a) opena communication link between the computer and positioner, (b) send commands to move thepositioner to its starting position which will be used as the origin reference. (3) Initialize thePNA to clear the PNA of its current settings and then load it with new settings. (4) Performthe measurement by: (a) move the positioners and acquiring frequency response data at all ofthe receiving grid points and (b) send the data from the PNA to the laptop and saving themfor future processing. (5) After data has been collected for all possible combinations oftransmitting and receiving grid points, end the program, close the PNA and reset the xy-positioner.80Figure 3-4 Flowchart of the UWB-MIMO channel sounder software81CHAPTER 4: EFFECT OF ANTENNA CONFIGURATIONON MIMO-BASED ACCESS POINTS IN ASHORT TUNNEL4.1 IntroductionAccess-point-to-access point (AP-AP) communication is needed in undergroundmines to extend wireless coverage to high traffic work areas of the mine such as the facearea. The face area is usually where the largest number of miners works and where thegreatest demand for wireless communication systems exists in underground mines. In recentyears, considerable effort has been devoted to characterizing conventional single-antennawireless channels in underground mine environments [113],[114]. These efforts havesupported replacement of legacy leaky feeder systems that provide basic voicecommunications by cellular telephony (in the range 0.8 GHz -2.1 GHz) and, most recently,by conventional wireless LAN technology (in the range 2.4 GHz -2.5 GHz) that provideintegrated voice and data communications. While the new systems perform well, furtherimprovements in coverage, throughput and reliability are required in order to support futureapplications such as mine automation and tele-operation of mining equipment.In this regard, multiple-input-multiple-output (MIMO), a compelling new technologythat multiplies data throughput, coverage and reliability without consuming extra transmitpower and radio frequency spectrum [6],[115] seems to be a perfect response to this demand.However, the mining industry is hesitant to adopt new wireless technologies such as MIMO,82as there are many uncertainties about the performance and deployment strategies. Theirconcern is valid because standardized off-the-shelf products, such as IEEE-802.11n systems(multiple-antenna technology), are designed and tested for conventional indoor and outdoorenvironments, and therefore are not necessarily suitable for underground mine tunnels withdifferent propagation behavior. In fact, the performance of such products is highly dependenton the radio channel which includes the surrounding propagation environment and theantenna configurations at both ends. This is even more evident for confined environmentswhere objects are located at closer distances to the antennas.A large number of experimental and analytical studies have been conducted todetermine the best antenna solution for conventional indoor and outdoor environments[18],[116],[117], resulting in various configuration designs for multiple-antenna systems.Conversely, not much research has been conducted on MIMO channel modeling and antennaconfiguration designs for underground tunnels, such as subways and underground mines,which have distinctive waveguide behavior. There are some theoretical MIMO studies basedon modal analysis, and limited number of experimental MIMO studies has also beenperformed for rectangular [24] and arched [21],[118],[119] long subway tunnels. Thesestudies mainly focus on the 900 MHz frequency band considering practical application ofGSM-R standard for railway communications and derived works in Europe. In [21], theauthors have studied and compared three array orientations and configurations with fixedspacing for AP to mobile communications, where a mobile antenna is installed on the trainwindshield. They have shown that in subway tunnels, array orientation impacts MIMOperformance, and within a certain distance from the Tx, MIMO capacity growth is achievablein spite of low angular spread imposed by the tunnel geometry [21].83While yielding useful insights about MIMO performance in long and large tunnels,these findings fail to predict the performance of MIMO systems operating in otherfrequencies and employing with other antenna orientations, configurations and locations thatmight be more convenient to be used for AP communications inside the underground minetunnels. More importantly, they cannot directly be applied to underground mine tunnels dueto their differences in size, branches, and extensive infrastructure they often have.A more recent study, which is theoretical and is based on modal analysis, aims tomaximize the capacity by optimizing only the number and position of MIMO antennas forunderground tunnels [120]. This has been done by allocating MIMO subchannels to highpower eigenmodes. The authors however, have not elaborated on practical considerations ofantenna array design, deployment and practical implications. The optimum position theyfound for the antenna is in the middle of the tunnel cross-section, which may not be practical.Moreover, this study is purely theoretical, is only for a straight tunnel, and therefore cannotcapture all the detailed geometry of an underground tunnel with extensive infrastructure.In addition to purely theoretical studies for general cases, we believe more realisticscenarios need to be studied using comprehensive simulation or measurement data thatcapture details of the propagation environment. Because conducting MIMO measurementsfor various antenna configuration scenarios is very time consuming and difficult due tolimited access to the underground mines, we chose to use ray tracing method to assessvarious configurations. Ray tracing simulation additionally allows: (1) assessment andcomparison of study cases which are difficult to be studied by measurement or theoreticalmodeling, e.g., the effect of infrastructure, (2) visualizing and obtaining physical insights byobserving the rays? interaction with the surrounding environment.84Accordingly, for this study, we chose to carefully simulate an underground servicetunnel which has the same size as of a typical mine tunnel with a long section (103.5 m), abranch, and extensive infrastructure. These features can more realistically characterizetypical underground mine tunnels while facilitate assessment of the research tool by allowingus to perform both simulation and measurement in an environment, where unlikeunderground mines, is easy to access. We used Wireless Insite ray tracing software (byRemcom Inc.) to assess the performance of the MIMO antenna configurations. Afterselecting the best performing configurations by ray tracing method, we have experimentallyvalidated the simulation results by conducting MIMO channel sounding in the service tunnel.The remainder of this chapter is organized as follows: In Sec. 2, we describe oursimulation setup and scenarios. In Sec. 3, multiple antenna analysis that has been used for ourstudy is given. In Sec. 4, we present simulation results and discussions. In Sec. 5, we validatesimulation results with measurement. Finally, in Sec. 6, we conclude the chapter bysummarizing our key findings and their implications.4.2 Simulation Setup and ScenariosIn this section, we describe the geometry of the site chosen for the study and alsoexplain how we carefully construct the service tunnel with its detailed structure in a ray-tracing program.4.2.1 Chosen SiteThe chosen underground tunnel located underneath the Woodward InstructionalResources Center at University of British Columbia, is a suitable site considering its size,ease of access, long and narrow pathways, and rich infrastructure. The entire structure of the85underground tunnel resembles a general cross-shape with a total length of 103.5 m and awidth of 55.7 m. The widths of the tunnel substructure vary from 2.7 m (main tunnel) to 6.8m (branched tunnel) with almost same height of 2.4 m throughout the entire structure (Figure4-1).To accurately model this tunnel, a 3D floor plan of the tunnel, the properties of all thestructures such as thickness, permittivity, conductivity, reflection, and transmissioncoefficients needed to be matched to the corresponding prototypes in the physical tunnel.Specifications for the materials used for walls and galvanized pipes are presented in Table 4-1. Sidewalls and ceiling were assumed to have same dielectric material, while metal doorswere considered perfect-electric-conductors (PEC).Figure 4-1 UBC Woodward service tunnel constructed in Wireless Insite with its extensiveinfrastructure and considering 3 propagation scenarios.86Table 4-1 Specifications of the main structures used in the Wireless Insite simulations.Electrical Properties?a (S/m) ?r Thickness (m)Concrete Walls 0.015 15 0.3Pipes 16 3.1 0.0034.2.2 Software and Parameter RestrictionTo accurately construct the floor plan with extensive rounded pipes, we found it moreconvenient to create the floor plan and infrastructure by making the tool file itself rather thanimporting them from AutoCAD or SolidWorks 3D software into the ray-tracing software. Todraw round pipes, we had to draw polygon patterns with the maximum allowable sides tomimic circular shape because WI cannot recognize the circular pattern. Also, to effectivelydraw the pipes with the appropriate side number, we had to find out the minimum lengthrecognizable by the ray-tracing tool.As a rule of thumb, the minimum length restriction is the wavelength of operationfrequency. Nevertheless, we performed a benchmark testing with different pipe edges to findthe exact cutoff geometry length before the infrastructure is not detectable by the ray-tracingtool; it was found to be 4 cm. As a result, all dimensions in the simulation, including thepolygon cross-sections of the pipes, were constructed with slightly larger edge length thanthe cutoff geometry length (4 cm) to closely resemble the original shape of the pipes whileremaining visible to the ray-tracing tool. Careful construction of the floor plan andinfrastructure made our simulation substantially accurate, however it required large memoryand long run time due to large number of facets interacting with the shot rays.874.2.3 Simulation ScenariosBased on preliminary results, we found that a 4?4-ULA (i.e., 4-element linear array atthe Tx and 4-element linear array at the Rx) configuration offers higher capacity compared to4?4-square one (i.e., 4-element square array at the Tx and 4-element square array at the Rx).Therefore, 4?4-square MIMO scenario was excluded and for further analysis, study wasnarrowed down to 4?4-ULA configurations. For our study, we considered typicalpropagation scenarios which are more common in underground mines such as: (1) pure LOSlink between the Tx and Rx inside a long tunnel, (2) LOS link with a branch in the middle,and (3) NLOS link with the Tx in the main tunnel and the Rx in the branched tunnel. Thesethree propagation scenarios are illustrated in Figure 4-1.Among various antenna configurations and deployments, we have chosen the onesthat are practical and convenient to be used in underground mines. Accordingly, MIMO APantennas are either placed close to the sidewalls or under the ceiling. The distance of theantennas from ceiling and sidewalls were kept more than a wavelength (about 12.5 cm). Foreach propagation scenario, we have considered several antenna configurations (differing inantenna polarization and array orientation) and deployments, which are summarized inTable 4-2. Antenna radiation pattern is assumed to be omnidirectional at the Tx and Rx.For array orientation parallel to the tunnel width, ?x?, parallel to the tunnel height,?y?, and parallel to the tunnel axis, ?long? (or longitude), have been chosen. For Rx1 andRx2 grids, ?x?, ?y?, and ?long? belong to the main tunnel, while for the Rx3 grid theybelong to the branched tunnel. For vertical polarization, antennas are placed parallel to thetunnel height, while for horizontal polarization they are placed parallel to the tunnel width.88Except for the deployment scenario 4 that has been considered at Rx1 grid location only,other six scenarios have been studied at all three Rx grid locations (Rx1, Rx2, and Rx3).Table 4-2 Different antenna configuration and deployment scenarios (x: parallel to the tunnel width,y: parallel to the tunnel height, long: parallel to the tunnel axis, H: horizontal polarizationand V: vertical polarization).Sc1(long-V)Tx, Rx:oppositesidewallsSc2(long-H)Tx, Rx:oppositesidewallsSc3(tr-y-V)Tx, Rx:oppositesidewallsSc4(tr-y-Hsmaewall)Tx, Rx:same wallonly at Rx1Sc5(tr-y-H)Tx, Rx:oppositesidewallsSc6(tr-x-V)Tx, Rx:underceilingSc7(tr-x-H)Tx, Rx:underceilingTx antennapolarization V H V H H V HRx antennapolarization V H V H H V HTx Arrayorientation long long y y y x xRx Arrayorientation long long y y y x xTo make it easier to visualize the antenna scenarios, Figure 4-2 is presented. In allcases, there is a grid of antennas at the Tx (5?3) and a grid of antennas at the Rx (5?5).Depending on the scenario under study, different Tx and Rx array orientations from the gridshave been chosen during the post processing. From the output files of ray tracing simulations,we have constructed sixty 4?4-H matrices for each antenna configuration.Due to interference from wireless LAN transmissions from upper floors, the 2.4-2.483 GHz frequency band could not be used for the MIMO measurement in the servicetunnel, therefore, 2.49 GHz was considered. We set the transmit power to 0 dBm and numberof received rays by each antenna element at receiver to 50. The upper limit number ofreflections and diffractions in each propagation path is set to 30 (maximum value) and 0,respectively.  Number of diffractions was set to zero because our preliminary results showed89that including diffraction does not significantly change the results, while dramaticallyincreases the computation time.Figure 4-2 Antenna configuration scenarios used in this study (V: vertical, H: horizontal).4.3 Multiple Antenna AnalysisIn this section, formulas and analysis that have been used for the multiple antennastudy are presented.4.3.1 Minimum Interelement SeparationTo determine suitable interelement separation, the correlation coefficient ofneighboring elements at the Rx was studied. Because the highest correlations correspond tosuccessive elements, we chose Pearson?s correlation coefficient of successive elements (m, n)at Rx grids for every Tx element (p) as follows:? ? ? ?? ?? ?hnphmphnpnphmpmpnpmphnphmphhEhhCorr ????? ???? ,,(1)90where hmp and hnp are channel gain matrix entries corresponding to mp and np subchannels,and also ? and ? represent mean and standard deviations, respectively.4.3.2 Normalization FactorWhile working on MIMO systems, one of the difficulties is that channel capacitydepends on the scaling of the H-matrix. Because in design of MIMO antennas or codingschemes absolute value of the scaling factor is not necessary, common practice is tonormalize H-matrices so that the average SNR at the receiver elements is set to a fixed valueand can be adjusted as a parameter. In order to perform a fair comparison of differentsystems or schemes, it is required to consider the same normalization factor sometimes. Here,two normalization factors are considered to compare performance of the antennaconfigurations, depending on whether the objective is to study: (1) the effect of subchannels?correlation only, or (2) both correlation and power impact on the capacity. For normalization1, every H-matrix is normalized by its own Frobenius norm which is the RMS value of theelements of a matrix and calculated as follows [6]:??? ???Rx TxNiNjijFFRxTxhNN1 122norHHHH(2)where H, NRx, NTx and hij are channel coefficient matrix, number of Rx antennas, number ofTx antennas, and channel matrix entries, respectively. Normalized H-matrices obtained bythis method are only affected by the level of correlation experienced at the antennas.Normalization 2 is often used for fair comparison of different MIMO antennaconfigurations [24],[121]. This normalization method, which is also called global channel91normalization [122] not only includes subchannels? correlation but also their powercontribution, which is different for each antenna configuration. For this normalizationmethod, a common normalization factor is considered for all scenarios under comparison,which is calculated based on averaging Frobenius norms of H-matrices of all cases beingcompared. Considering K total channel realizations at each transmitter location, thisnormalization can be calculated as follows:???? ? ???KkNiNjijRxTxRx TxhNKN1 1 122norHHHH(3)After finding normalized H-matrices, capacity C (in bit/s/Hz) of a MIMO systemwith NTx transmitting antennas and NRx receive antennas can be calculated by [6]:???????????????? ?? *nornor HHITxN NSNRCRxdetlog 2 (4)where ? denotes the transpose-conjugate, Hnor is the NRx?NTx normalized channel matrix,SNR is average signal to noise ratio and I is identity matrix. It is assumed that the NTx sourceshave equal power and are uncorrelated. CDFs of channel capacity for different scenarioshave been obtained assuming average SNR=20 dB.4.4 Simulation Results and DiscussionsIn this section, simulation results are presented, discussed, and accompanied byphysical interpretation. Based on the results, antenna interelement separation and antennaconfigurations that offer high performance are identified. To study the impact of tunnel92infrastructure, the results are compared for the tunnel with and without the infrastructure. Theoutcome of this section can be useful in developing guidelines to design and deploy MIMOarray configurations in underground mines.4.4.1 Antenna Interelement SeparationTo quantify the correlation among the MIMO subchannels, which determines suitableinterelement separation, we computed the correlation coefficient of successive elementsbased on equation (1) for different interelement separations at the Rx for different scenarios(with and without infrastructure). We have tabulated the result for horizontal array withvertically polarized antenna in Table 4-3.From Table 4-3, we can see: (1) ?/2 separation which is a typical antenna separationfor most MIMO off-the-shelf products shows high correlation and (2) correlation betweensuccessive elements is higher for with infrastructure case compared to without infrastructurecase. Observation (2) is relatively unexpected, because one might expect to see decrease inthe level of subchannels? correlation after including the infrastructure. Our simulations haveshown that although infrastructure inside a tunnel may increase the number of multipaths, itincreases correlation of neighboring elements, and degrades channel capacity (as we will seein the following section).Both ray tracing and waveguide modelings can explain this result. Considering theray tracing modeling, this can be due to the fact that key rays between Tx and Rx, whichcontribute in forming decorrelated subchannels have been blocked by the infrastructure. Onthe other hand, considering the waveguide theory this result can be attributed to thewaveguide mode suppression that has occurred due to the presence of the infrastructure. This93decreases decorrelation of orthogonal waveguide modes present in an empty rectangulartunnel. Based on Table 4-3, by choosing antenna separation of 2?, correlation of less than 0.5can be achieved for both cases of the tunnel with and without infrastructure, and therefore,for further analysis we chose 2? interelement separation.4.4.2 Array Orientation and Antenna PolarizationIn this section, the objective is to assess and compare the antenna configurations withdifferent array orientations and antenna polarizations for different propagation scenarios inthe tunnel with and without infrastructure, and also to select two antenna configurationswhich show closest capacity CDF to that of an i.i.d. Rayleigh fading channel. Here, to obtainand compare channel capacity for different scenarios, power impact of each antennaconfiguration is included. This has been performed by normalizing all the H-matrices at eachRx grid location with a common normalization factor (global normalization).Table 4-3 Correlation coefficient of neighboring elements at Rx with different interelementseparations (V: vertical, H: horizontal polarizations).?/2 ? 3?/2 2?WithInfrastructureLOSV 0.69 0.35 0.23 0.24H 0.83 0.58 0.34 0.13NLOSV 0.80 0.63 0.57 0.49H 0.76 0.55 0.45 0.39WithoutInfrastructureLOSV 0.47 -0.09 -0.22 -0.19H 0.82 0.51 0.14 -0.18NLOSV 0.51 0.09 0.03 0.13H 0.40 0.01 0.04 0.10944.4.2.1 Tunnel without InfrastructureFigure 4-3 compares the antenna configurations for different propagation scenarios inthe empty service tunnel. As the results show, the antenna configurations perform verydifferently, confirming significance of antenna configuration design in linear confinedspaces. The different performance is particularly more evident in Rx1 scenario wherearriving rays are more concentrated compared to the other scenarios (Rx2 and Rx3).As can be seen from Figure 4-3, the tr-x-V and tr-y-H configurations at Rx1, and tr-x-V configurations at Rx2 and Rx3 shows the closest capacity CDF to that of an i.i.d. Rayleighfading channel. It is interesting to note that based on the waveguide theory, these two antennaconfiguration would be equivalent if: (1) the tunnel aspect ratio is equal to one (i.e., widthand height have the same size), (2) the material of all the walls is the same, and (3) thedistance of the arrays from the wall (or ceiling) is equal for each scenario (tr-y-H and tr-x-V).In our study, however, tunnel width is larger than the tunnel height, and also arrays? distancefrom the wall (or ceiling) is not the same. Due to practical constraints, array distance fromthe ceiling in tr-x-V scenario is larger than array distance from the sidewall in tr-y-Hscenario.Higher performance of tr-x-V configuration compared to tr-y-H configuration can bedue to two main reasons. First, the tr-x-V array is placed closer to the center of the cross-sectional plane. When the antenna is placed closer to the center of the cross-sectional plane,energy coupling from the antenna to the waveguide will be stronger. Therefore, tr-x-Vconfiguration has higher power compared to tr-y-H configuration. Second, tr-x-Vconfiguration has larger angular span (in azimuth plane, ?-plane) compared to tr-y-Hconfiguration (in elevation plane, ?-plane), because the tunnel width is larger than the tunnel95height. This can result in lower correlation among the MIMO subchannels in tr-x-Vconfiguration.Comparing tr-y-H(same wall) and tr-y-H (or tr-y-H(oppwall)) deployments at Rx1 showthat although strongest rays in tr-y-H(same wall) scenario has higher power compared  to the tr-y-H, as can be seen from Figure 4-3, tr-y-H configuration shows higher capacity.Observation of the rays explains this result by showing that the Tx and Rx arrays in tr-y-Hscenario are more exposed to each other and the walls (due to their relative position), andthus larger number of rays can be generated between the Tx and Rx in tr-y-H5(oppwall).  Backwalls and in some cases infrastructure can also increase the Tx and Rx interaction.At Rx2, tr-x-V shows significantly better performance compared to the tr-y-Hconfiguration, which is not what one may expect considering the impact of the branch in themiddle of the transmitter and Rx2. The branch allows some of the propagation rays in themain tunnel to escape into the branch tunnel. This can be even more severe for configurationswith vertically polarized antennas that mainly use sidewalls to propagate. Results at Rx3confirm this fact by showing higher channel capacity for the tr-x-V configuration and twoother configurations with vertically polarized antennas (long-V and tr-y-V).The reason for high capacity at Rx2 found to be the positive impact of the sidewalls?indentions at Rx2 location as can be seen in Figure 4-4. The sidewalls? indentions work infavour of the tr-x-V configuration by causing waveguide mode conversion, which improvesthe channel capacity. Figure 4-4 compares propagation rays for the tr-x-V and tr-y-Hconfigurations at Rx2. As it illustrates for the tr-y-H configuration, multiple reflectionsformed by the back wall are not sufficiently strong (due to the large distance between the Rxand the back wall) to be able to compete with the tr-x-V configuration.960 5 10 15 20 25 3000.51CDF of the 4x4 MIMO Capacity (No infrastructure)0 5 10 15 20 25 3000.51CDFSc1_long-VSc2_long-HSc3_tr-y-VSc4_tr-y-H_same wallSc5_tr-y-HSc6_tr-x-VSc7_tr-x-HRayleigh_MIMORayleigh_SISO0 5 10 15 20 25 3000.51Rx2Rx3Rx1Channel Capacity (bit/sec/Hz)Figure 4-3 Comparison of different antenna configurations for the tunnel without infrastructure.Figure 4-4 Comparison of the performance of tr-x-V and tr-y-H antenna configurations at Rx2.974.4.2.2 Tunnel with InfrastructureFigure 4-5 compares the antenna configurations for different propagation scenarios inthe service tunnel including the infrastructure. The results suggest that infrastructure impactsthe performance of each antenna configuration differently. For most cases, the infrastructureacts as a waveguide mode suppressor and degrades the channel capacity by blocking the keyrays between the Tx and Rx. The worst impact can be seen on the tr-x-V configuration. Theconfiguration that we found to be the best performing one for the tunnel withoutinfrastructure shows the worst performance for the tunnel with the extensive infrastructure.On the other hand, the tr-y-H configuration offers the best performance and the closestcapacity CDF to that of an i.i.d. Rayleigh fading channel for all three propagation scenarios.0 5 10 15 20 25 3000.51CDF of the 4x4 MIMO Capacity  (With Infrastructure)0 5 10 15 20 25 3000.51CDF0 5 10 15 20 25 3000.51Sc1_long-VSc2_long-HSc3_tr-y-VSc4_tr-y-H_same wallSc5_tr-y-HSc6_tr-x-VSc7_tr-x-HRayleigh_MIMORayleigh_SISORx3Channel Capacity (bit/sec/Hz)Rx1Rx2Figure 4-5 Comparison of different antenna configurations for the tunnel with infrastructure.98We can also see degradation for the configurations with vertically polarized antennasdue to particular impact of infrastructure on vertically polarized rays. In an empty tunnel, weobserved that vertically polarized waves reflected from the sidewalls are much stronger thanthe ones reflected from the ceiling/floor (unlike horizontally polarized waves). This can beexplained by considering the fact that vertically polarized waves undergo lower attenuationwhen reflecting from vertical walls (i.e., sidewalls) in an oversized dielectric waveguide[123]. Additionally, the null of the antenna radiation pattern is towards the ceiling/floorwhile its maximum is toward the sidewalls. The infrastructure in the service tunnel, however,does not allow formation of strong specular reflections from the sidewalls.As can be seen in Figure 4-6, in the Tx1-Rx1 propagation scenario where so manymasts are located close to one of the sidewalls (Figure 4-6), antenna configurations such astr-x-V with vertically polarized antennas do not perform as suitable as horizontally polarizedones. This is the main reason why the tr-x-V configuration which performs best in the emptytunnel shows the worst performance at Rx1. Figure 4-6 clearly indicates the impact ofinfrastructure on its performance.  From Figure 4-6, we can see that when the tunnel isempty, several strong rays that are comparable to the LOS path travel to the Rx. However,when the infrastructure exists only direct path (if it exists) is strong while the ones thatundergone multiple reflections from the infrastructure are very weak (i.e., lower order modesuppression). This can result in two effects which both degrade the capacity: (1) lowerreceived power and (2) large LOS path compared to other multipaths generated mostly bymultiple reflections from the infrastructure.99Figure 4-6 Impact of infrastructure on performance of tr-x-V-Sc6 antenna configuration.Because the tr-x-V, tr-x-H, and tr-y-H configurations are the better performingconfigurations in different scenarios, we chose them for further analysis and comparison oftheir power. Table 4-4 is presented to show the effect of infrastructure on each of them.Depending on the MIMO array configuration and Rx location in the tunnel, the infrastructureimpacts pathloss differently. Unlike all other cases, in the tr-y-H case at Rx2, theinfrastructure showed positive impact which can be due to the additional reflections formedby the infrastructure between the Tx and Rx. Table 4-4 also reveals that tr-y-H configurationoffers the highest and tr-x-V the lowest power level in the service tunnel with extensiveinfrastructure.Based on both Table 4-3 and Table 4-4, for most scenarios infrastructure increasesboth the correlation and the power loss of the subchannels, which consequently degrades theMIMO channel capacity. As it can be seen, in spite of the extensive infrastructure for mostcases pathloss is less than free-space-path-loss (FSPL).100Table 4-4 Average pathloss at 2.49 GHz.WithoutInfrastructureWithInfrastructureFree Space Pathloss(2.49 GHz)PLave (dB) PLave (dB) FSPL (dB)Rx1dTx-Rx=31.3 mSc4(tr-y-H-same wall) 62 64 70Sc5 (tr-y-H) 54 58 70Sc6 (tr-x-V) 54 71 70Sc7 (tr-x-H) 57 60 70Rx2dTx-Rx=16 mSc5 (tr-y-H) 57 54 64Sc6 (tr-x-V) 46 58 64Sc7 (tr-x-H) 50 54 64Rx3dTx-Rx=7.6 mSc5 (tr-y-H) 65 74 58Sc6 (tr-x-V) 52 78 58Sc7 (tr-x-H) 62 76 584.4.3 Main Observations and Conclusions from Ray Tracing SimulationsAlthough the results obtained from assessing the antenna configurations based on theray tracing simulations may seem to be site-specific, they provide physical insight fordeployment strategies. Main conclusions of this study can be summarized as follows:? Three waveguide mechanisms can be considered which govern the propagation in thetunnel as an oversized dielectric waveguide: a) mode coupling from the antenna to thetunnel (depending on the antenna location in the cross-sectional plane and the antennapolarization), b) mode coupling between waveguide modes (may be caused by branch,back walls, walls? indentions, etc.), and c) waveguide mode suppression (may be causedby infrastructure, curvature, etc.).? Array configuration and antenna polarization can significantly impact MIMO systemperformance in short underground tunnels.101? Interelement spacing of 2? between antenna elements at both ends (Tx and Rx) providessufficient decorrelation.? As the simulation results reveal, LOS deployments of MIMO antenna configurationsoffer more desirable performance in the short tunnel, particularly when the infrastructureis extensive, which results in additional power loss. Therefore, deployments which ensureLOS link as well as larger interactions between the Tx and Rx are preferred.? However, in the LOS cases, MIMO capacity seems to be more affected by the choice ofantenna configuration and deployment compared to the NLOS case. Therefore, antennaconfiguration requires more careful design in LOS cases than in NLOS cases.? Because rays are more concentrated for scenarios at Rx1(where Tx and Rx are located inthe long section of the tunnel) than the ones at Rx2 and Rx3, antenna configurationdesign requires more attention.? Two antenna configurations of tr-x-V and tr-y-H have been identified as the bestperforming antenna configurations for the empty tunnel and the tunnel withinfrastructure, respectively. While tr-x-V configuration remarkably outperforms all otherconfigurations in the empty tunnel, it shows very poor performance in the case ofextensive infrastructure.? While infrastructure was found to be a degrading factor for most cases, some geometricalproperties of tunnel such as sidewalls? indentions show positive impact on the MIMOsystem performance.1024.5 Experimental ValidationFrom results obtained by the ray tracing simulations, two better performing MIMOarray configurations were chosen for experimental validation. Employing the MIMO channelsounder, we experimentally validated their performance predictions. More detail about theMIMO channel sounder we used is given in Table 4-5. We placed the Tx and Rx grids in theexact same locations as we had in the ray tracing simulations. At the Rx side, an xy-positioner (3?5 points) was used whereas at the Tx a single antenna was being moved at sixpoints.Table 4-5 Measurement specificationsUnit SpecificationVNA master Anritsu MS2034AFrequency bandNumber of frequency points2.49 - 4 GHz551Antenna Electrometrics UWB Biconical antennasPositionerTx powerVelmex linear xy-positioner26 dBmGrid point spacing 2? at 2.49 GHzDynamic rangePower amplifierFibre optics120 dBOphir - Model 5303075Miteq SMCT-100M11GExperimental and simulation results of capacity statistics for tr-x-V and tr-y-Hconfigurations are presented in Table 4-6. Global normalization factors at each Rx gridlocation for with and without infrastructure have been used (in total six normalizationfactors). Six normalization factors are calculated so that fair comparison can be performedbetween different antenna configurations at each Rx grid. Measured H-matrices have alsotheir normalization factors at each Rx grid location (three normalization factors for threeRxs). As can be seen from Table 4-6, simulation and the measurement results show good103agreement (except for the tr-x-V configuration at Rx1).  In addition to tabular presentation,simulation and measurement results for two antenna configurations of tr-x-V and tr-y-H arecompared in Figure 4-7. Similarly, the experimental results show good agreement withearlier simulation results and confirm higher performance of tr-y-H configuration comparedto tr-x-V configuration for the tunnel with extensive infrastructure.Table 4-6 Capacity statistics at 2.49 GHz and SNR=20 dB to compare MIMO capacity of differentantenna configurations (with power-impact)With Infrastructure Measurement (With Infrastructure)Cave ? Coutage%10 Cave ? Coutage%10Rx1Sc4(tr-y-H-same wall) 15.68 1.66 13.82 16.20 1.62 14.11Sc5(tr-y-H) 21.86 1.92 19.7 22.93 2.03 19.99Sc6(tr-x-V) 5.96 0.59 5.34 11.59 1.32 10.01Rx2Sc5(tr-y-H) 20.67 2.69 17.66 21.28 1.98 18.86Sc6(tr-x-V) 13.36 2.43 10.63 16.40 1.62 14.36Rx3Sc5(tr-y-H) 21.03 2.08 18.67 21.98 2.16 19.15Sc6(tr-x-V) 16.68 2.52 13.03 17.96 1.76 15.59Although this figure shows a relatively good agreement between the simulation andmeasurement, it indicates that the ray tracing simulation underestimates the capacity for bothantenna configurations. This underestimation is more noticeable for tr-x-V configuration inthe pure LOS scenario (Rx1) where rays are more concentrated compared to other scenarios(Rx2 and Rx3). This implies that the accuracy of the ray tracing method is not consistent fordifferent scenarios (i.e., Rx grid locations). At low capacities which can be associated tohighly concentrated rays, achieving same results from both measurement and simulation isless likely due to higher sensitivity of the results to small errors in the modeling. Forexample, the results of ray tracing may be very sensitive to the array locations, and a slight104difference between exact array locations in the measurement and simulation may result insignificant errors.As can be seen from Figure 4-7, the deviation of simulation and measurement is lessfor Rx2 and Rx3 scenarios where rays are more spread. This implies that ray tracingmodeling for the tunnels might not be as accurate as it is for conventional indoorenvironments, therefore further enhancement for modeling the tunnels is required. This issuemay be resolved by increasing the ray resolution (by allocating larger number of rays perangle unit) for the study areas with highly concentrated rays. But in the case of extensiveinfrastructure, runtime will substantially increase.0 5 10 15 20 2500.51CDF of the 4x4 MIMO Channel Capacity at SNR=20 dB0 5 10 15 20 2500.510 5 10 15 20 2500.51Channel Capacity (b/sec/Hz)Sc5_tr-y-H_MeasSc6_tr-x-V_MeasSc5_tr-y-H_SimSc6_tr-x-V_SimRx2Rx3Rx1Figure 4-7 Comparison of simulated (with infrastructure) and measured MIMO capacity CDFs.105The MIMO measurements in the service tunnel allowed us to obtain more accurateresults, however it did not provide any physical insights. Whereas by conducting ray tracingsimulations, we could observe and trace the propagation rays, and obtain physical insights,but it failed to offer consistent accuracy for different scenarios.4.6 ConclusionsIn this study, we have shown that in spite of low angular spread in undergroundmines, careful design of antenna configuration allows benefiting from MIMO technology forAP-AP communications. Our results reveal that array orientation, antenna separation, andpolarization significantly influence performance of multiple-antenna systems in undergroundtunnels. Consequently, capacity can be improved by careful antenna configuration design inunderground mines.We have determined sufficient interelement separation, antenna polarization, andarray orientation. Antenna separation of ?/2 that is common in most of the off-the-shelfproducts, does not show high performance. Our study showed MIMO antenna separationrequired to be 4 times longer than in conventional indoor environment with rich multipath(i.e., 2?).We compared different configurations based on their capacity, sensitivity to typicallocations in a mine and sensitivity to the infrastructure. In this study, we also examined theeffect of geometry of the tunnel such as indentations of the walls and infrastructure.Infrastructure found to be a degrading factor on capacity for most cases due to blockages itcauses.106We have identified two practical MIMO access point configurations which showbetter capacity (bit/sec/Hz) for a 4?4-MIMO system inside an underground service tunnel:horizontal array with vertically polarized antennas (tr-x-V) and vertical array withhorizontally polarized antennas (tr-y-H). While the tr-x-V configuration performs remarkablybetter in different propagation scenarios in the tunnel with little or no infrastructure, ifextensive infrastructure exists, the tr-y-H configuration which was found to be less affectedby the infrastructure, performs significantly better than tr-x-V configuration. As a result,infrastructure density should be considered as an important factor while designing MIMOantenna configuration for deployment in linear confined spaces such as underground tunnels.107CHAPTER 5: EFFECT OF ARRAY PROPERTIES ONMIMO SYSTEM PERFORMANCE IN ANUNDERGROUND MINE5.1 IntroductionMIMO systems are a well-proven wireless technology for use in surfaceenvironments where they offer higher data rates, greater coverage and increased reliabilityfor line-of-sight (LOS) and non-line-of-sight (NLOS) scenarios compared to oldertechnology. Nevertheless, their performance is uncertain in confined spaces such asunderground mines. The rapid growth of MIMO-based technologies (e.g., reconfigurableantennas, advanced space-time coding schemes, multi-user MIMO) in the past decade, hasmade it more crucial to resolve physical layer limitations when deploying MIMO systemsinto new environments to maximize performance.Currently, most underground mines are equipped with legacy communication systemssuch as leaky feeders that suffer from limited coverage, low data rates, and require an LOSpath. These limitations make MIMO technology an attractive solution for wirelesscommunications inside mines because MIMO has overcome these problems in surfaceenvironments. However, MIMO-based systems are highly dependent on the surroundingphysical environment. Therefore, conventional MIMO-based wireless devices that work wellin surface environments do not necessarily work well in underground mine tunnels wherepropagating signals are characterized by a low angular spread.108Unlike the extensive MIMO studies conducted for indoor and outdoor environments,only a few experimental studies have been conducted for underground environment such assubway tunnels [21]- [24] and underground mines [124]. In [124], as the only experimentalMIMO study in underground mines, performance of a 2?2-MIMO system for two types ofantenna radiation patterns has been compared in a very short underground mine (25 m).However, further investigation of antenna properties and the effect of the mine structure(e.g., curvature) on the channel capacity are required. Relevant studies for subway tunnelshave focused on access-point (AP) to mobile communications at 900 MHz [21], in whichtransmitter antennas are on the platform and receiver antennas are located on the trainwindshield. While previous studies confirm that MIMO can be a promising technology forAP-to-mobile applications in underground environments, too many uncertainties still remain.For example, there is no performance prediction, array design, deployment strategy orchannel prediction available for MIMO-based AP communications inside undergroundmines. In addition, due to geometric dissimilarities between mines and subway tunnels, theapplicability of the results obtained for tunnels to mines, is questionable.In response to these uncertainties, we have conducted both theoretical andexperimental MIMO performance analysis for underground mines. We have evaluatedMIMO-based AP communications and answered some of the questions left in the literature.We have also shown how antenna properties (e.g., antenna spacing, polarization and height)impact wireless performance, and thus should be carefully considered in array configurationdesigns and deployments inside mines.We have experimentally evaluated the MIMO system performance in an actual minesite, and study the impact of other antenna properties, such as height and polarization. By109employing different normalization methods on the channel coefficient matrix (H-matrix), wehave differentiated the ways that antenna properties impact the capacity (the spatial structureor the power). We have also studied the effect of mine curvature on the MIMO channelcapacity.We have compared the experimental results in the mine tunnel with a theoreticalmodel based on the waveguide theory (multimode waveguide model) [70], which wasdeveloped for underground tunnels and claims to be applicable to underground mine tunnels.Our experimental work matches well with this model, and confirms its applicability tounderground mine tunnels. We have used this model to achieve physical insights and explainour experimental results. Additionally, this model allows us to evaluate the channel capacityof several antenna spacings and accordingly determine the proper antenna spacing forMIMO-based systems.Finally, we conclude this study by providing results that can be useful in developingguidelines for MIMO system deployments in underground mines. The remainder of thischapter is organized as follows. In Sec. 2, we describe the measurement site, our setup andmeasurement scenarios. In Sec. 3, the multiple antenna analysis used in this study ispresented. In Sec. 4, results and impact of antenna properties, which can be used in MIMOsystem deployment in underground mines are discussed. Finally, in Sec. 5, we conclude thechapter by presenting key findings and their implications.1105.2 MIMO Measurement in an Underground Mine5.2.1 Measurement Site and EquipmentWe performed a MIMO measurement campaign at Myra-Falls underground minelocated in Strathcona Park on Vancouver Island, B.C., Canada (Figure 5-1). The mine gallerywas 1900 ft (579 m) below the surface. The cross-sectional shape of the tunnel wasrectangular. The width and height of the tunnels varied from 5m to 5.7 m and from 3.5 m to4.2 m, respectively. The mine?s floor was covered in mud, and the only infrastructure wasleaky feeder cables, wires and pipes installed along the ceiling. The sidewalls were notperfectly straight, and had several protrusions, but their roughness was insignificant. We hadthe opportunity to conduct static measurements because the mine-level was closed andnobody was working there. Photography of MIMO measurement setup at Myra Falls is givenin Figure 5-2. More details on our measurement setup and calibration of the equipment havebeen described in Chapter 3.For this study, we used our MIMO ultra-wide-band (UWB) channel sounder system.It uses the virtual array method, in which mutual coupling effect of the antenna elements isnot included in the channel characterization. Although this channel sounder performs UWBmeasurements, the results and analysis presented in this work focuses on MIMOcharacterization at frequencies near the 2.4 GHz, industrial-scientific-medical (ISM) band.We used an Anritsu MS2034A vector-network-analyser (VNA) to send and receivefrequency span of 2.49 GHz - 4 GHz with 551 frequency points. Two UWB Biconicalantennas were used, one at the transmitter and one at the receiver. The transmitter antennawas manually moved across the tunnel and was connected to the VNA using an RF-over-fibre range extender. The receiver antenna was automatically moved on a fixed xy-positioner111(1 m ? 0.5 m). Locations of the receiver (Rx) grid and transmitting (Tx) array in the mine aredemonstrated in Figure 5-1. For each transmitter position, the frequency dependent transferfunction between the transmitter antenna and all 15 virtual receivers was measured.Figure 5-1 Map of the Myra-Falls mine in B.C., Canada and the transmitting (Tx) array and receiver(Rx) grid locations (5 Tx array locations and 2 Rx grid locations).Figure 5-2 Photography of RF equipment in Myra Falls mine.1125.2.2 Measurement ScenariosIn spite of time constraints in the mine, spatial samples were collected from variousparts of the tunnel. Measurement scenarios for different propagation scenarios, such as LOS,NLOS after a curvature, etc. are as follows: short-distance LOS (10 m, Tx1-Rx1), long-distance LOS (27 m, Tx2-Rx1), curve shaped NLOS (49 m, Tx3-Rx1), NLOS in a branch(12 m, Tx4-Rx1) and LOS with a branch in the middle (21 m, Tx5-Rx2). All of thesemeasurements can be considered as typical scenarios in underground mines. As shown inFigure 5-1 and Table 5-1, five sets of measurements were taken: four Tx locations (Tx1-Tx4)for Rx1 and one Tx location (Tx5) for Rx2. At the Tx locations, 4 antenna positions with aseparation of 2 wavelengths (2?) (at 2.49 GHz) were considered in the middle of the tunnelcross-section. At the Rx locations, the virtual array was implemented by using a xy-positioner with 15 evenly spaced points. Location Rx1 was used for all but one of thescenarios, in which a branch was located in the middle of the direct Tx-to-Rx path.5.2.3 Measurement DesignBecause no previous work has determined the sufficient spacing required for multipleantenna measurements in underground mines, we investigated previous studies on similarlinear confined spaces to select the minimum separation. Several studies in hallways showthat the coherence distance is larger than ? [125],[19], and thus the common separation of ?/2for conventional indoor environments is not sufficient. In addition to hallways, weconsidered the results of our earlier ray-tracing study and development runs in a mid-sizeservice tunnel (width=2.7 m, height=2.4 m, length=103.5 m). A transmitter and a receivergrid with vertically polarized antennas were located about 35 m apart and in the middle of thetunnel. Because the highest correlations correspond to successive elements, we chose113Pearson?s correlation coefficient of successive elements (m, n) at the Rx grid for Tx antennaas follows:? ? ? ?? ?? ?hnhmhnnhmmnmhnhmhhEhhCorr ????? ???? ,,(1)where hm and hn are channel coefficients corresponding to m and n channels, and also ? and ?represent mean and standard deviations, respectively. The envelope correlation coefficientsfor two separations of ?/2 and 2? were found to be 0.82 and 0.18, respectively.  Based on thisresult and previous work in hallways, we chose 2? spacing between antenna elements for theTx array and the Rx grid.To see the impact of array height, we considered two heights of under the ceiling(2.7 m above ground) and medium height (1.7 m above ground), and to study the impact ofantenna polarization, for both vertical and horizontal antenna polarizations, measurementswere performed at medium height. In summary, we collected data for three antennascenarios: (1) ceiling height with vertically polarized antennas (CV configuration), (2)medium height for vertically polarized antennas (MV configuration), and (3) medium heightfor horizontally polarized antennas (MH configuration). For all the Rx grid points, weperformed frequency domain measurement and collected amplitude and phase of the channelgain over the frequency range of 2.49 GHz - 4 GHz. Measurement scenarios are summarizedin Table 5-1.To select the suitable array orientation, Pearson?s correlation of successive antennaelements was found for two orientations, one perpendicular and one parallel to the tunnelaxis. Figure 5-3 shows how successive antenna elements are chosen for each arrayorientation. Results of envelope correlation coefficients for all measurement scenarios are114summarized in Table 5-2. Correlation coefficient values consider all four Tx antennas (i.e.,Tx array) for each measurement scenario.Table 5-1 MIMO measurement scenarios in the Myra Falls underground mine.Tx1 Arrayat Loc.1Tx2 Arrayat Loc.2Tx3 Arrayat Loc.3Tx4 Arrayat Loc.4Tx5 Arrayat Loc.5Rx1GridLOS (10 m)MVLOS (27 m)(before a curve)MH,MV,CVNLOS (49 m)(after a curve)MH,MV,CVNLOS (12 m)(in a branch)MH,MV,CV----------Rx2Grid ---------- ---------- ---------- ----------LOS  (21 m)(a branch in the middle)CVFigure 5-3 Spatial correlation analysis on the successive antenna elements on the Rx grid for twodifferent array orientations: (a) perpendicular to the tunnel axis and (b) parallel to thetunnel axis.As it can be seen in Table 5-2, array orientation perpendicular to the tunnel axis offersmuch lower correlation compared to the array orientation parallel to the tunnel axis.Choosing array orientation to be perpendicular to the tunnel axis with its elements separatedby 2? provides a sufficient degree of decorrelation (less than 0.7) regardless of the antennapolarization, height, and propagation scenario. Intuitively, this orientation provides suitabledecorrelation as it occupies the largest space orthogonal to the arriving signals. Therefore, we115chose this array orientation with element separation of 2? for constructing 4?4-MIMO H-matrices. While such considerations for array design have not been evaluated forunderground mines, the results of capacity analysis will determine whether this design isappropriate or not.Table 5-2 Envelope correlation coefficients of Rx grid antennas for different measurement scenarios(f=2.49 GHz).Rx Array Orientation Perpendicularto the Tunnel?s AxisRx Array Orientation Parallelthe Tunnel?s AxisMH MV CV MH MV CVLoc. 1LOS (10 m) ---------- 0.24 ---------- ---------- 0.40 ----------Loc. 2LOS (27 m)Before Curvature0.21 0.10 0.63 0.67 0.75 0.86Loc.3NLOS (49 m)After Curvature0.30 0.46 0.07 0.83 0.76 0.80Loc.4NLOS (12 m) 0.40 0.07 0.27 0.47 0.54 0.59Loc.5LOS (21 m)Branch-Middle---------- ---------- 0.45 ---------- ---------- 0.915.3 Multiple Antenna Analysis5.3.1 Constructing Channel Coefficient MatricesWe started MIMO analysis, after ensuring the absence of large-scale fading across thechosen Tx array. This was examined by averaging the channel power over all the frequenciesand Rx grid points on the positioner for each Tx location and comparing them all. Then, asshown in Figure 5-1, six 4?4-MIMO spatial realizations based on uniform-linear-arrays(ULA) oriented perpendicular to the tunnel axis, are constructed (from the 4-element Tx116array and 3?5-element Rx grid). In addition to spatial realizations, 551 frequency samplesover range of 2.49-4 GHz and separated by 2.74 MHz were collected.For our analysis, we consider a bandwidth of 96 MHz (35 frequency samples).However, because frequency samples within one coherence-bandwidth (BWc) are correlated,all 35 samples cannot be treated as independent. Adding frequency samples to our spatialones requires them to be independent, which implies separation of at least one BWc betweenthem [121]. Coherence bandwidth was found to be 4 MHz from the channel-frequency-response. Therefore, only half (17 samples) of the 35 frequency samples (with separation of2.74 MHz) can be considered independent. As a result, 102 (17?6) 4?4-MIMO H-matriceswere constructed at each location.5.3.2 H-Matrix Normalization and Channel CapacityIt is a common practice in MIMO antenna design to normalize H-matrices so that theaverage SNR at the receiver elements is set to a fixed value and can be adjusted as aparameter. Depending on the objective of MIMO analysis, different methods can be used tocalculate normalization factor. In order to perform a fair comparison of different systems orschemes, it is required to consider one normalization factor for all scenarios under the study.Here, two the normalization factors are considered for performance comparison of differentantenna scenarios. The choice of normalization factor depends on whether the objective is tostudy the effects of:  (1) subchannels? correlation only, or (2) both correlation and power, onthe capacity.For normalization (1), every H-matrix is normalized by its own Frobenius norm,which is the RMS value of the elements of a matrix and calculated as follows [6]:117??? ???Rx TxNiNjijFFRxTxhNN1 122norHHHH(2)where ||.||F denotes Frobenius norm and H, NRx, NTx, and hij are channel coefficient matrix,number of Rx antennas, number of Tx antennas, and channel matrix entries, respectively.Normalized H-matrices obtained by this method are only affected by the level of correlationexperienced by the antennas.Normalization (2) is often used for fair comparison between different MIMOscenarios (antenna polarization, array height, etc.) or to study the effect of something (e.g.,curvature) on the MIMO performance [24],[121]. This normalization method, which is alsocalled global channel normalization [122], not only includes subchannels? correlation butalso power contribution of each case. For this normalization method, a commonnormalization factor is considered for all scenarios under comparison, which is calculatedbased on averaging Frobenius norms of H-matrices of all cases being compared. ConsideringK total channel realizations at each transmitter location, this normalization can be calculatedas follows:???? ? ???KkNiNjijRxTxRx TxhNKN1 1 122norHHHH(3)Note that in both methods, the effect of the antenna gain (Biconical at 2.49 GHz) andthe pathloss between the transmitter and receiver have been excluded. After finding118normalized H-matrices, capacity C (bit/s/Hz) of a MIMO system with NTx transmittingantennas and NRx receive antennas can be calculated by [6]:???????????????? ?? *nornor HHITxN NSNRCRxdetlog 2 (4)where ? denotes the transpose-conjugate, Hnor is the NRx?NTx normalized channel matrix,SNR is average signal to noise ratio and I is identity matrix. It is assumed that the NTx sourceshave equal power and are uncorrelated.5.4 Measurement Results and Discussions5.4.1 Performance Comparison of MIMO Antenna ScenariosChannel capacity which is the main performance measure of MIMO systems, isinfluenced by two factors: subchannels? power and the level of spatial correlation amongsubchannels. Therefore, to study the effect of array properties on the MIMO channelcapacity, we chose to determine how each array property impacts these two factors. First, weassess them based on their impact on spatial correlation without considering theircontribution in power, and afterwards we include the power aspect too. This can be done byapplying different normalization methods to the H-matrices. Channel Capacity Without Power ConsiderationsIn this section, channel capacity CDFs have been found for different measurementscenarios and the pathloss between the centres of the Tx array and Rx grid is calculated,while the antenna element gain and array?s power impact (differences due to the array heightand antenna polarization) at each Rx grid are excluded. Therefore, the degree of subchannels?119decorrelation is the only factor that controls the capacity. This has been done by normalizingevery H-matrix by its own Frobenius norm (normalization (1) ). Figure 5-4 presents thecapacity CDFs of measured channels alongside the CDF of an i.i.d. Rayleigh fading channelto compare the measured channels to an ideal channel, in terms of MIMO performance.12 14 16 18 20 22 24 2600. CDF of different scenarios with no power considerations (SNR=20dB )Loc1 (MV)Rayleigh12 14 16 18 20 22 24 2600. (CV)Loc2 (MH)Loc2 (MV)Rayleigh12 14 16 18 20 22 24 2600. (CV)Loc4 (MH)Loc4 (MV)Rayleigh12 14 16 18 20 22 24 2600. (CV)Loc3 (MH)Loc3 (MV)Rayleigh12 14 16 18 20 22 24 2600. Capacity (bit/sec/Hz)Loc5 (CV)RayleighFigure 5-4 CDFs of 4?4-MIMO capacity without power considerations (based on the measurement).As Figure 5-4 shows, the capacity CDFs for different antenna heights andpolarizations are similar and close to that of an i.i.d. Rayleigh fading channel. This impliesthat sufficient spacing and properly chosen array orientation provide a sufficient degree ofdecorrelation regardless of the antenna polarization and height. However, further analysis is120required to see whether they impact subchannels? power. We can also see that the spatialdecorrelation achieved by the proposed array is not sensitive to the propagation scenario inthe mine.Another way of analysing the spatial structure of MIMO subchannels is to studysingular values of the H-matrices. Average singular values corresponding to differentmeasurement scenarios in the mine are presented in Table 5-3. All singular values arenormalized to the largest singular value. Table 5-3 also confirms that spatial structure of thewireless channel is very close to that of an i.i.d. Rayleigh fading channel (rich-multipathchannel), and provides required decorrelation among the subchannels. Therefore, the spatialstructure of the wireless channel is suitable and has the potential for achieving multiplexinggain. The rich-multipath for close Tx-Rx distances has also been reported in other studies[126],[127]. In [127], several amplitude distributions, such as Nakagami, Gamma, Rice,Rayleigh and Lognormal, are tested to model small-scale fading. Using a Kolmogorov-Smirnov?s (KS) goodness-of-fit test, the authors have shown that the measured LOS scenario(1 m-12 m) can be modeled by both Nakagami and Rayleigh distributions.To provide a physical explanation and justification for observing decorrelated MIMOsubchannels in the short underground mine tunnel, we have employed the multimodewaveguide model, which has been developed and used by Sun et al. [70] to characterizewireless propagation in underground tunnels.  In this model, a tunnel is considered to be anoversized dielectric waveguide and the modes obtained by the waveguide theory are allpossible solutions of Maxwell?s equations that can exist in the tunnel. We also used thismodel for MIMO channel characterization and evaluation of our experimental results in the121underground mine tunnel. More detail, including the mathematical expressions of this modelfor tunnels with rectangular cross-sectional shape, can be found in [70].Table 5-3 Singular values of measured H-matrices and i.i.d. Rayleigh H-matrices (all singularvalues are normalized to the largest one).Average Singular Values?1 ?2 ?3 ?4i.i.d. Rayleigh fading channel 1 0.69 0.40 0.14Loc. 1(10 m)MV 1 0.71 0.39 0.14Loc. 2(27 m)CV 1 0.55 0.34 0.15MH 1 0.67 0.39 0.15MV 1 0.63 0.33 0.11Loc. 3(49 m)CV 1 0.70 0.38 0.14MH 1 0.67 0.39 0.16MV 1 0.55 0.30 0.10Loc. 4(12 m)CV 1 0.67 0.36 0.13MH 1 0.69 0.40 0.14MV 1 0.70 0.44 0.15Loc. 5(21 m) CV1 0.61 0.32 0.12Angular spread is a key indicator that shows whether a wireless propagationenvironment has the potential of offering spatially decorrelated subchannels or not. Largerangular spread offers higher spatial decorrelation (or equivalently lower correlation) amongthe multiple antennas. To characterize the angular spread, we applied the multimodewaveguide model to an equivalent rectangular tunnel with the same cross-section as the MyraFalls mine (width: 5.5 m and height: 4 m). Angular spread in a tunnel can be found asfollows [128]:122/2 /22/2 /2/2 /2/2 /2( ) ( )  ( )( ) ( )rmsA d A dA d A d? ?? ?? ?? ?? ? ? ? ? ? ?? ?? ? ? ?? ?? ?? ?? ??? ?? ?? ?(5)where ?rms and A(?) are the azimuth angular spread and power azimuth spectrum,respectively. Figure 5-5 shows that the angular spread of the near zone area (including 10 mdistance) is quite large. This large angular spread in the near zone is the result of strongreflections from the walls. For further distances (more than about 200 m), angular spreadbecomes very low (about 4o) due to high attenuation of higher order waveguide modes [70].To improve spatial decorrelation among subchannels at further distances, antenna separationmay need to be increased (more than 2?).10 50 100 150 200 250 300 350 400 450 50041024Angular spread variation vs. distanceDistance (m)Angular Spread (deg)Figure 5-5 Angular spread variation versus distance (based on multimode waveguide model).To exploit the potential of the spatial decorrelation offered by the surroundingenvironment and achieve the multiplexing gain, interelement spacing of the antenna arrayshould be chosen properly. Figure 5-6 (a) shows the capacity as a function of distance, andFigure 5-6 (b) shows capacity CDFs for different antenna separations obtained by using themultimode waveguide model. As can be seen from both figures, the common antenna123spacing of the off-the-shelf products (?/2) is not sufficient, and does not offer consistentperformance over Tx-Rx distance. On the other hand, 2? separation shows suitable andconsistent performance (capacity), and spacing the antennas further than that (e.g., 6?) doesnot achieve higher capacity. This confirms our measurement results. For very large Tx-Rxdistances however, the element spacing may require to be increased. The capacity CDF for2? spacing also matches well with the capacity results measured in the mine.8 10 12 14 16 18 20 22 24 26 2700. Capacity (bit/sec/Hz)(b)Capacity CDF for Myra Falls tunnel at 2.4 GHzCDF10 20 30 40 50 60 70 80 90 1001015202530Capacity for Myra Falls tunnel at 2.4 GHz (SNR=20 dB)Distance (m)(a)Capacity (b/Hz/sec)6?2??/2Figure 5-6 Capacity of 4?4-MIMO system for different antenna spacings (based on the multimodewaveguide model). Channel Capacity With Power ConsiderationsTo include the power impact of array height and antenna polarization, Tx-Rx pathlossand antenna element gain have been removed from the H-matrices by considering a commonnormalization factor, normalization (2), for all the configurations at each Rx grid. In this124way, capacity is calculated so that it includes the impact of antenna polarization and arrayheight on both subchannels? decorrelation and power. Figure 5-7 shows capacity CDFs fordifferent antenna scenarios at different measurement locations. At Loc. 1 and Loc. 5, oneantenna configuration was considered while for the rest of locations three configurationswere measured. Comparing all the figures, the largest difference among CDFs of antennaconfigurations is about 5 bit/sec/Hz in after the curvature scenario and between MH/CV withMV configuration.12 14 16 18 20 22 24 2600. 14 16 18 20 22 24 2600. (CV)Loc2 (MH)Loc2 (MV)Rayleigh12 14 16 18 20 22 24 2600. (CV)Loc3 (MH)Loc3 (MV)Rayleigh12 14 16 18 20 22 24 2600. (CV)Loc4 (MH)Loc4 (MV)Rayleigh12 14 16 18 20 22 24 2600. (MV)RayleighLoc5 (CV)RayleighChannel Capacity (bit/sec/Hz)Capacity CDF of different scenarios with power considerations (SNR=20dB)Figure 5-7 CDFs of 4?4-MIMO capacity with power considerations (based on the measurement).5.4.2 Impact of Antenna PolarizationFigure 5-7 shows similar performance for MH and MV scenarios, however, The MHscenario has a slightly better capacity CDF (about 1 bit/sec/Hz) for both Loc. 2 (LOS) and125Loc. 4 (NLOS). This similar performance of two polarizations is likely due to the highdepolarization in the near zone of the tunnel. However, because the aspect ratio of the tunnelis horizontal (i.e., tunnel width is larger than tunnel height), horizontal polarization shows abit lower attenuation in the near zone. Based on our results, this is only valid for the straightpart of the tunnel. In Loc. 3, where a curvature exists between the transmitter and thereceiver, MV shows the best performance by a large margin (5 bit/sec/Hz) for both MH andCV cases. Poor performance of MH in this case, can be attributed to the fact that horizontallypolarized waves are attenuated more than vertical ones by the curvature, and thus thedegradation of capacity can be due to the power loss.Similar behaviour has been reported for horizontally polarized waves propagating in asubway tunnel, which is substantially larger than a typical underground mine [13]. Assume arectangular tunnel as shown in Figure 5-8. Equations (6) and (7) give the attenuation of EHmnmode (?mn) in this tunnel for y-polarization (vertical polarization) and x-polarization(horizontal polarization), respectively [13],[123]:????????????????????????????????1Re2211Re22 22hhwymn hnhwmw ?????? (6)????????????????????????????????11Re221Re22 22hwwxmn hnhwmw ?????? (7)where w, h, ?, ?w and ?h are tunnel width, tunnel height, wavelength, relative permittivity ofvertical walls (sidewalls) and relative permittivity of horizontal walls (ceiling/floor),respectively. As can be seen from (7), vertical walls (sidewalls) on which horizontallypolarized E is perpendicular to, contribute to most of the attenuation [123]. In [123], it isshown that if E is perpendicular to the curved walls, more attenuation occurs due to126additional loss caused by the curvature. In fact, curvature of the sidewalls can be consideredas a polarization filter that passes only vertically polarized waves, and its level of filtering isinversely proportional to the radius of the curvature.Figure 5-8 A tunnel with rectangular cross-section.5.4.3 Impact of Array HeightFor all the Rx locations with CV scenario, we can see that the ceiling height showslower capacity compared to the middle height. Capacity median is found to be 2-5 bit/sec/Hzless at the ceiling height compared to the middle one (MV). Based on the waveguide theory,antenna located in the middle of the tunnel can couple higher power to the tunnel due toexcitation and reception of the dominant propagation mode.  Although it may not be practicalto place the antenna array in the middle of the tunnel, this result suggests placing the antennafurther from the ceiling for short underground tunnels. If this is not possible due to practicalconsiderations, reduction on capacity should be taken into account while designing MIMOsystems. This result may not necessarily be valid for very long Tx-Rx distances.  Over verylong distances (several hundred metres or more), only a limited number of waveguidepropagation modes exist because the higher order modes are significantly attenuated. Higherorder modes, acting like multipath components are an important factor to construct spatiallydecorrelated MIMO subchannels. Therefore, to excite higher order modes with sufficient127energy to remain active for long distances, off-centered position of Tx and Rx arrays in thetransverse plane may be preferred over centered position, as described in [24]. This impliesthat at far distances a trade-off may be required between subchannels? power anddecorrelation.To evaluate the results obtained in this section, Table 5-4 compares the mean pathlossbetween each Tx antenna and Rx grid for each measurement scenario. Free space pathlosswhich is calculated considering a longitudinal distance between the center of the transmitterarray and center of the Rx grid, is also given. As it can be seen, MH shows the lowestpathloss at Loc. 2 and Loc. 4, while showing the highest one at Loc. 3 (after the curvature).We can also see relatively similar performance for both polarizations at Loc. 2 and Loc. 4,and better performance for the middle height compared to the ceiling height. These resultsmatch the results obtained from Figure 5-5, and thus confirm previous discussion on differentpower contribution of different array configurations.Experimental results of capacity predictions with power considerations are comparedwith results from the multimode waveguide model for two LOS cases at 10 m and 27 m inthe straight part of the mine. Figure 5-9 (a) and Figure 5-9 (b)-(d) compare capacity CDFsobtained by measurement and multimode model for Loc. 1 and Loc. 2, respectively. InFigure 5-9 (b)-(d), capacity CDFs from the multimode waveguide model for different heightsand polarizations are obtained based on normalization (2). The experimental results matchwell with the theoretical results obtained for both locations, heights, and polarizations. Thisindicates that the multimode waveguide model can be considered as a simulation tool toeffectively design and deploy advanced systems such as MIMO-based devices in minetunnels.128Table 5-4 Mean pathloss measured at each Rx grid location in Myra-Falls mine (at 2.49 GHz).Mean Pathloss (dB)Free SpacePathloss (dB)Loc.1LOS (10 m)MVTx1_antenna1 -62.31-60.36Tx1_antenna2 -60.42Tx1_antenna3 -59.64Tx1_antenna4 -60.52CV MV MHLoc.2LOS (27 m)Before CurvatureCV,MH,MVTx2_antenna1 -69.84 -68.19 -67.71-69.00Tx2_antenna2 -69.71 -68.46 -68.54Tx2_antenna3 -71.43 -69.89 -68.85Tx2_antenna4 -71.24 -70.51 -68.73Loc.3NLOS (49 m)After CurvatureCV,MH,MVTx3_antenna1 -83.19 -78.62 -83.56-74.17Tx3_antenna2 -81.66 -79.67 -82.04Tx3_antenna3 -82.14 -77.59 -81.02Tx3_antenna4 -83.08 -76.93 -82.00Loc.4NLOS (12 m)CV,MH,MVTx4_antenna1 -80.30 -78.18 -77.79-62.00Tx4_antenna2 -80.54 -78.63 -76.75Tx4_antenna3 -80.05 -78.34 -77.64Tx4_antenna4 -80.22 -79.57 -78.44Loc.5LOS (21 m)Branch-MiddleCVTx5_antenna1 -72.65-66.82Tx5_antenna2 -70.97Tx5_antenna3 -71.55Tx5_antenna4 -68.0812912 14 16 18 20 22 24 26 2800.51Loc2 (MH) - MeasuredLoc2 (MH) - MultimodeRayleigh12 14 16 18 20 22 24 26 2800.51CDF12 14 16 18 20 22 24 26 2800.51Capacity (b/sec/Hz)Loc2 (MV) - MeasuredLoc2 (MV) - MultimodeRayleighLoc2 (CV) - MeasuredLoc2 (CV) - MultimodeRayleigh14 16 18 20 22 24 26 2800.51 Capacity CDF of Myra Falls main tunnelLoc1 (MV) - MeasuredLoc1 (MV) - MultimodeRayleighd=10 md=27 md=27 md=27 m(a)(d)(b)(c)Figure 5-9 CDFs of 4?4-MIMO capacity with power considerations (based on measurement andmultimode waveguide model at f=2.49 GHz, SNR=20 dB).5.5 ConclusionsBy this study, we showed that antenna array properties such as array orientation,height, antenna spacing, and polarization greatly impact the performance of MIMO systemsin underground mines. Our experimental and theoretical analysis revealed that deploying 4-element ULAs with element separation of 2?, horizontal orientation, and placedperpendicular to the tunnel axis provides a sufficient degree of decorrelation amongsubchannels to achieve a suitable MIMO capacity (close to that of an i.i.d. Rayleigh fadingchannel) in several propagation scenarios studied in a short underground mine tunnel.130We showed that if MIMO antenna arrays are deployed based on the proposed arrayorientation and interelement separation, array height and antenna polarization will mainlyimpact the power of the MIMO subchannels without significantly changing their spatialstructure. This implies that by employing the proposed array design, one can choose otherantenna properties (e.g., antenna polarization, height) while devoting main focus on poweraspects (similar to SISO systems) rather than concerning about the spatial correlation amongthe MIMO subchannels.We also confirmed previous studies by showing that the mine geometry is a keyfactor that should be taken into account while designing wireless systems for undergroundtunnels. As an example, although in mines with a horizontal aspect ratio,  horizontallypolarized antennas show lower attenuation, and thus are more desirable, if a curvature existsbetween the transmitter and receiver, vertical polarization is preferred as it is much lessaffected by the curvature loss.In spite of the dissimilarities between underground mines and tunnels, some of ourresults confirm and support results obtained in subway tunnels. Nevertheless, becauseunderground mines are geometrically diverse, more measurement campaigns in variousmines are required to reconfirm previous findings and reveal new physical trends orprinciples.131CHAPTER 6: CHARACTERIZATION OF ANGULARSPREAD IN UNDERGROUND TUNNELSBASED ON A MULTIMODE WAVEGUIDEMODEL6.1 IntroductionResearch on wireless channel modeling allows the underground mining industry tosave time in their wireless system deployments. Underground tunnels and mines have beenintensively studied, both theoretically and experimentally, since the late 1970?s. The mainfocus has been on the pathloss and delay spread, which are parameters of interest in single-input single-output (SISO) systems. For characterization of MIMO channels, however, thespatial domain becomes as important as the temporal domain. Similar to the power-delay-profile (PDP), the power-azimuth-spectrum (PAS) has been defined, which determines thespatial distribution of the received power in azimuth. Consequently, angular-spread (AS) isdefined as the standard deviation of PAS, being equivalent to the root-mean-square (RMS)delay-spread of PDP [8]. AS is found to predict many properties of MIMO-based systems,e.g., singular value and capacity distributions [9].Because the AS is expected to be very different in mine tunnels compared toconventional indoor and outdoor environments, employing off-the-shelf MIMO products thatare primarily designed for conventional environments may lead to poor performance inunderground tunnels. As a result, conducting MIMO channel modeling and characterizing132AS allow suitable guidelines for designing MIMO antenna configurations customized forunderground tunnels to be developed. Accordingly, deployment parameters such asseparation of the antenna elements, antenna polarization, array arrangement, and transverseposition can be determined.For surface environments, AS and antenna configuration designs have beenextensively studied for more than a decade. However, MIMO research, and in particular,antenna configuration design and characterization of angular spread for wireless propagationin tunnels and mines, is sparse, and related work in literature is limited. Recently, AS wasexperimentally studied in [128] for a large arched subway tunnel (8.6 m?6.1 m) with slightlyrough walls (i.e., roughness is less than 2 cm). However, no further investigations on angularspread characterization, such as finding a distribution fit for PAS inside a tunnel has beenperformed in previous studies.In our study, we have used a theoretical model, multimode waveguide [70] andperformed more comprehensive angular spread characterization. This theoretical model waschosen due to its accuracy and capability to model the near zone of tunnels. As previousstudies [70],[75] show, for upper ultra-high-frequency (UHF) frequencies in tunnels, themultimode waveguide model is more accurate than its counterparts such as the single-modewaveguide model proposed by Emslie et al. [11], and dual-slope pathloss models (orbreakpoint model) proposed by Zhang [114]. Additionally, unlike the single-modewaveguide model, multimode models includes higher order modes, and is therefore capableof characterizing both near and far zones of a tunnel.Employing this model, we characterized the PAS, AS, and correlation coefficientobserved by antenna elements within different zones in a tunnel, including close distances133(10 m- 50 m). These angular properties are found to be dependent on both the tunnel size andthe tunnel zone, i.e., the transmitter-receiver separation. We studied and compared them fortwo tunnel sizes and three zones inside the tunnel. Without loss of generality, we focused onshort tunnels (less than 500 m) in this chapter. The same methodology can be applied tolonger tunnels. The results of AS characterization can be used to extend the IEEE-802.11nMIMO channel model to underground mines or as the basis for other correlation-basedMIMO channel models for underground tunnels. In addition to angular properties, we havecharacterized channel capacity for different interelement spacings and tunnel zones in orderto evaluate the improvement in multiplexing gain that can be achieved in such confinedspaces. We compared and validated our findings to experimental work presented in [128],and also to simulations using a ray-tracing based software (Wireless InSite from RemcomInc.) [129].The remainder of this chapter is organized as follows. In Sec. 2, the multimode modelis introduced. In Sec. 3, assessment of angular properties and channel capacity based on themultimode waveguide model is presented. In Sec. 4, results and model validation arediscussed. Finally, in Sec. 5 we summarize our key findings and contributions.6.2 A Multimode Waveguide ModelMultimode modeling has been developed and used by Sun et al. [70] to characterizewireless propagation in tunnels.  In this model, a tunnel is considered to be an oversizeddielectric waveguide. Here, we assume that the x-axis is along the tunnel?s width, the y-axisis along its height and the z-axis is along its length. Considering a y-directed (verticallypolarized) dipole as a source of excitation, the field distribution of y-polarized hybrid modes134at any position (x, y, z) inside the tunnel, can be found by solving Maxwell?s equations forthe specified cross-sectional dimensions and the material of the tunnel in terms ofeigenfunctions as follows [70],[130]:? ? ?????? ??????? ?? yxeigennm yhnxwmyxE ???? sincos,,(1)where w and h are width and height of the tunnel, respectively, and ?x and ?y are paritydesignations that take values of 0 or ?/2 depending on whether the integer values of m and nare even or odd [130]. It is assumed that the co-polarized field components play a dominantrole and the cross-polarized components can be ignored. The intensity of each mode dependson the excitation, which can be found by applying a geometrical-optical (GO) model in theexcitation plane (transmitting plane), i.e., the cross-sectional plane that contains thetransmitting (Tx) antenna. The electromagnetic (EM) field distribution in the transmittingplane is, in fact, the weighted sum of the field of all modes. The mode intensities are thenestimated by a mode matching technique as given in (2) [70] assuming that transmitter isvertically polarized (y-polarized) and located at (x0, y0, z0).? ? ? ? ? ? 0,),(,,,1 100zzjeigennmm nmnRx mnmneyxEyxCzyxE ?????????? ?? (2)where Cmn, ?mn and ?mn are the mode intensity on the excitation plane (which depends on thetransmitter location), attenuation coefficient and phase shift coefficient, respectively.We have employed this model in our characterization of the angular properties andMIMO channel capacity in underground tunnels. Verification of the multimode model withexperimental work performed in [13] is given in Figure 6-1. For this verification, both135antennas were vertically polarized and placed at a height of 2-m at the same horizontallocation at a distance 1/4 of the tunnel width from the tunnel sidewall.0 500 1000 1500 2000 2500-120-100-80-60-40-200Normalized power for a large size tunnelTx-Rx Distance (m)E-field Power (dBm)ExperimentMultimode 450 MHz900 MHzFigure 6-1 Comparison of multimode model with experimental work obtained from [13].6.3 Angular Spread and Capacity Characterization Based on aMultimode Waveguide Model6.3.1 Angular Spread Characterization Based on a Multimode WaveguideModelIn this section, we use the multimode waveguide model to characterize angular spreadin a rectangular tunnel. We calculate the azimuth angle of the propagating rays inside atunnel. Then, we find the PAS and determine the correlation coefficient of antenna elementsin a uniform-linear-array (ULA). Power Angular SpectrumTo find the distribution of received signal strength versus angle-of-arrival (AOA) inboth the azimuthal and elevation planes, we use equations for a rectangular waveguide.136Azimuth AOA (AOA-?) and elevation AOA (AOA-?) of each propagation mode (m, n) canbe calculated using the following equations [131]:?????????hnwmnm2)sin(2)sin(????(3)where ?, ?, ? are the azimuth angle, elevation angle and wavelength, respectively. Azimuthangles of two waveguide modes that use sidewalls to propagate are illustrated in Figure 6-2.Angle (? or ?) is a function of mode index (m or n), transverse dimension of the tunnel (w orh) and operation frequency. Their power is additionally dependent on the positions oftransmitter and receiver, and can be found from 2RxE , using (2) at each receiver point (x,y,z).Figure 6-2 Reflection angles of lower and higher order modes from sidewalls in a waveguide (?1 is theangle for the lower order and ?2 is for the higher order mode). Correlation Coefficient of Antenna ElementsSpace selective fading caused by AS makes signal amplitude dependent on thelocation of the antenna [10]. The parameter that characterizes space selective fading is thecoherence-distance (DC) that is inversely proportional to the AS. It is defined here as thespatial separation for which the autocorrelation coefficient of the spatial fading drops to 0.7.In general, for a ULA with d separation between antenna elements, Pearson?s correlation canbe obtained as follows [132],[133]:137? ? ? ?? ? ? ? ? ?? ? ? ? ? ????????????????????????????????dPASDDRdPASDDRdDDjRDRXYXXXYXXsinsinsincos2(4)where ?, ?, ?, and PAS are the wavelength, correlation coefficient, angle of arrival, andpower-azimuth-spectrum, respectively.  Using (4), the envelope correlation |?| vs. antennaelement separation (d/?) can be plotted, which provides useful information for antenna arraydesign.6.3.2 H-Matrix and MIMO Channel CapacityTo construct MIMO matrices (H), formula (1) was substituted into (2) to find thechannel coefficients as follows:? ? ? ? jimnmn zzjm niieigennmjjjmnRxTxij eyxEyxCkGGh ??????????1 1,,),(2?? (5)where i is receiver index, j is transmitter index, hij is complex channel gain between ithreceiver and jth transmitter, GTx and GRx are transmitter and receiver antenna gains, k is thewave number, jmnC is the mode intensity of jth transmitter and , ( , )eigenm n i iE x y is  the value of (m,n) mode eigenfunction at ith receiver. After constructing H-matrices, capacity (in bit/sec/Hz)that is a fundamental property of MIMO-based systems can be calculated as follows [6]:???????????????? ?? *2 detlog HHITxN NSNRCRxbit/sec/Hz (6)where * denotes conjugate transpose, NTx is the number of transmitter antennas, NRx is thenumber of receiver antennas, H, hij and I are the normalized channel coefficient matrix,138channel matrix entries and identity matrix, respectively. H-matrices are normalized by theirown Frobenius norms, and therefore, are only affected by the level of correlation experiencedat the antennas. This formula is used to find the channel capacity distribution and variation ofcapacity vs. distance.6.4 Results and Validation of Multimode Modeling6.4.1 Simulation ScenariosIn our simulation setups, we considered omni-directional antennas with verticalpolarization for both transmitter and receiver, which are located in the middle of the tunnelunder study. While most results are obtained for a small tunnel with a typical cross-sectionalsize of underground mines, we have also considered a larger tunnel size (subway tunnel) bothwith the same length of 500 m. Two tunnel sizes were considered to: 1) compare our workwith a previously published work in which wireless propagation in a subway tunnel wasstudied, and 2) study the impact of the tunnel size on the angular properties.The two tunnels have following dimensions: 1) a small rectangular mine tunnel with3.8 m height and 5.1 m width (dimensions of an underground mine at Myra Falls onVancouver Island), and 2) a large subway tunnel with semi-circular cross-section that isexperimentally studied in [128]. As shown in Figure 6-3, the diameter of the cylindrical partof the subway tunnel is 8.6 m, the maximum height is 6.1 m at the centre of the tunnel andthe roughness of the walls is quite low, in the order of less than 2 cm (0.16?). A rectangularequivalent of this tunnel is shown in Figure 6-3. To validate our angular spreadcharacterization based on the multimode model, we compared our results to results obtainedusing a ray-tracing tool [129] and to experimental work presented in [128].139Figure 6-3 The cross-section of the subway tunnel and its equivalent rectangle.6.4.2 Characterization of PAS and AS6.4.2.1 Simulation SetupTo find the PAS in a tunnel, first based on the tunnel size and operating frequency(2.4 GHz), propagating modes and their corresponding angles can be found. Then, the powerof modes, which is dependent on the Tx and Rx locations should be found. Therefore, atransmitter and a grid of receivers are located at the middle height (y=0). The transmitter islocated at (0,0,0) and the receiver grid is considered over the area of ????????4:2:4wwx ? ,? ?mmmz 500:1:10? . To find the PAS at each receiver point, the power of each propagatingmode has been calculated. To plot the AS at each longitudinal distance (along the z-axis), theAS was averaged over the tunnel width (along the x-axis).After finding the power of corresponding azimuth angles, by using an iterative least-mean-square (LMS) distribution fitting, the PAS for different zones has been estimated andmodeled by a Gaussian function as follows:140? ?22121 ???????? ??? ?????? ??? ep (7)where ?, ?? and ?? are azimuth angle, azimuth AS and mean AOA, respectively. By findingthe standard deviation (i.e., AS) and mean of the Gaussian function, the Gaussian fit can beplotted. Figure 6-4 shows Gaussian distribution fitted to PAS obtained by multimodesimulation for the small tunnel.-80 -60 -40 -20 0 20 40 60 8000. fit of power azimuth spectrum for the small tunnelAngle of Arrival (deg)Normalized PowerMultimodeGaussian FitFigure 6-4 Zero-mean Gaussian fit of the PAS for the small tunnel (multimode waveguide model).Graphs in Figure 6-5 show Gaussian fits to the PAS for different zones of the smalland large tunnels. The Gaussian fits have zero-means, because the direct path between thetransmitter and the receiver is taken as the reference. Similar trends can be seen for bothtunnels and their corresponding zones; the further the distance, the smaller the AS. Thelargest difference can be seen in the near zone, and the smallest is observed in the far zone.For the tunnel with larger cross-section, AS is larger, however, in the far zone impact of thecross-section is less, and both tunnels show very similar AS. Larger AS in the near zone isdue to the fact that more higher order modes are active in this region.141-80 -60 -40 -20 0 20 40 60 8000. fit of power azimuth spectrumNormalized Powerd=10-50m, ?=15.3?d=50-150m, ?=7.7?d=150-500m, ?=4.6?-80 -60 -40 -20 0 20 40 60 8000. of Arrival (deg)d=10-50m, ?=19.9?d=50-150m, ?=8.8?d=150-500m, ?=4.8?Large TunnelSmall TunnelFigure 6-5 Zero-mean Gaussian fit of the PAS for different zones of two tunnel sizes (multimodewaveguide model). Validation by the Ray-Tracing MethodUsing the ray-tracing tool, similar results and same trend have been observed for thescenarios studied by the multimode model. Figure 6-6 compares the PAS for the entire smalltunnel obtained by both methods. As can be seen from this figure, two curves match quitewell but the angular spreads are slightly different. More detailed results are summarized andcompared in Table 6-1. Both methods show similar values in different tunnel zones;however, angular spreads obtained by the ray-tracing method are slightly larger than the onesobtained by the multimode model. Based on these results, PAS can be modeled by atruncated-Gaussian function with zero-mean and AS between 4.5o and 23o, depending on thetunnel zone and transverse dimension.142-80 -60 -40 -20 0 20 40 60 8000. azimuth spectrum for the small tunnelAzimuth Angle (deg)Normalized PowerRay-tracing, ?=7.0?Multimode, ?=6.0?Figure 6-6 Comparison of PAS obtained by multimode and ray-tracing for the small tunnel (wholetunnel).Table 6-1 Angular spreads (deg) for different zones of the large and small tunnel obtained bymultimode and ray-tracing methods.Large Tunnel(Subway)Small Tunnel(Underground mine)Multimode Ray-tracing Multimode Ray-tracingClose Distance10 m-50 m 19.9o 22.3o 15.3o 16.6oMiddle Distance50 m-150 m 8.8o 9.7o 7.7o 7.6oFar Distance150 m-500 m 4.8o 6.9o 4.6o 6.0oWhole Tunnel10 m-500 m 6.9o 7.7o 6.0o 7.0o6.4.3 Characterization of AS vs. DistanceFigure 6-7 shows AS vs. Tx-Rx distance for the small and large tunnels. As can beseen in the near zone of both tunnels, AS is a decreasing function of distance, and in the farzone, it remains constant at a small value (4o). Unlike lower order modes that arrive withgrazing angles, higher order modes undergo large number of bounces and are greatlyattenuated, thus they do not contribute much in AS at long distances. Therefore, AS143decreases as distance increases. This small AS, which is much smaller than AS ofconventional indoor environments (i.e., 12o-40o) [132],[134] requires different MIMOantenna configuration design for underground tunnels than are used in conventionalenvironments.Higher order modes are stronger in the near zone of larger tunnels, and therefore,their AS is expected to be larger in that area. This is confirmed by the results presented inFigure 6-7. Over the span 10 m-150 m, the impact of tunnel size on the AS is large. But afterabout 150 m, it gradually becomes insignificant and both tunnels show almost the same AS,and finally, it becomes independent of the tunnel transverse dimension.10 50 100 150 200 250 300 350 400 450 5004102030Angular spread vs. distance for different size tunnelsDistance (m)Angular Spread (deg) Large TunnelSmall TunnelFigure 6-7 AS variation over distance for two tunnel sizes (multimode waveguide model). Validation of AS vs. Distance by Ray-Tracing MethodFigure 6-8 compares three methods for characterizing AS vs. Tx-Rx distance for thelarge subway tunnel (8 m?5.6 m). As it can be seen, all methods similarly show the sametrend of rapid decrease before the breakpoint and slow decrease afterwards. Unlike theexperimental result, which captures constructive and destructive interactions among the rays,144both multimode and ray-tracing models characterize AS as a monotonically decreasingfunction of distance. In addition, both slightly overestimate the AS for a measured channel inan arched tunnel [128]. This deviation from the experimental results can be attributed to theshape of the actual tunnel (i.e., not being rectangular) and its walls (i.e., not being entirelysmooth). The antenna directivity can also be another impacting factor. In fact, the accuracyof measurement-based characterization of angular properties is highly dependent on theantenna directivity.50 100 150 200 250 300 350 400 450 500241012.515Angular spread vs. distance at 2.4 GHz - different methodsDistance (m)Angular Spread (deg) MultimodeRay tracingExperiment-4.5 ?/100m2.6?Figure 6-8 Comparison of three methods for characterizing AS in a large tunnel (ray-tracing andexperiment graphs are obtained from [128]).In Figure 6-8, we can also identify a breakpoint at a distance of about 200 m fromtransmitter, which distinguishes two zones of the tunnel. The first zone can be modeled by aregression line with a slope of 4.5o/100 m, and the second zone by a regression line with azero slope at 2.6o (mean value of AS) that remains constant for the entire region. Byincluding the near zone of 10 m-50 m in multimode and ray-tracing simulations, a morecomplete picture of angular properties can be obtained compared to previous study presentedin Figure 6-8. Based on both methods, in this zone, AS is large and close to that of rich145scattering indoor environments, however, as the distance increases it dramatically drops witha much sharper slope compared to other zones. Therefore, three zones can be identified forthis tunnel: (1) 10 m- 50 m, with very sharp decrease of AS, (2) 50 m-150 m, with mediumdecrease of AS, and (3) 150 m-500 m, with constant AS. Comparison with Other EnvironmentsAs was shown in Figures 6-8 and 6-9, AS is found to be a decreasing function ofdistance in tunnels based on different methods employed in our study and in [128]. Unlikethis consistent results, showing AS dependency on the distance in tunnels, for outdoorenvironments conflicting results have been reported on distance-dependency of AS[8],[135],[136]. Based on the literature, PAS has been characterized and modeled as uniform,truncated-Gaussian and truncated-Laplacian distributions [8],[136]-[138] for different indoorand outdoor scenarios. Local features that can vary by antenna height [8], and alsodistribution of scatterers around the receiver and transmitter have been found to be keyfactors in determining angular properties of a propagation environment.10 50 100 150 200 250 300 350 400 450 5000510152025303540Distance (m)Angular Spread (deg)Angular spread vs. distance - Large tunnelRay-tracing (Wireless InSite)MultimodeRay-tracingExperimentFigure 6-9 Comparison of different methods for characterizing AS in a large tunnel, including 10 m-50 m zone (ray-tracing and experiment graphs are obtained from [128]).146In Table 6-2, the angular properties of several environments are summarized forcomparison purposes. For outdoor environments, in most scenarios of both urban and rural,measured PAS has been fitted to a Laplacian function [8] for some cases Gaussiandistribution has also been obtained [8],[138]. For rural areas, AS at the base station is verysmall and about 2o. In macro-cellular urban situations, depending on the antenna height,median AS varies between 5o to 10o (larger for shorter height) [8]. For indoor environments,PAS can be modeled by Laplacian or Gaussian distribution with AS=12o in line-of-sight(LOS) scenarios, and a uniform distribution in non-LOS scenarios (if distribution ofscatterers is uniform) [8]. Although, tunnels are confined spaces and have small AS, ourresults show larger AS compared to macro-cellular base stations in rural areas (AS=2o), inparticular in near zone.  In near zone, AS is close to that of indoor but in the far zone, it is farless than typical AS in indoor environments.Table 6-2 Comparison of AS for different environments.PAS AS Ref.IndoorUniform,Gaussian,Laplacian12o-40o [137]UrbanMostlyLaplacian(some cases:Uniformand Gaussian)5o-10o [8],[136],[139]Rural Laplacian 2o [7]Tunnels Gaussian Near zone=15o-23oFar zone=4.5o-5o Our studyFor any PAS model chosen for a wireless propagation environment, a physicaljustification can be given. For example, physical explanation for a Laplacian distributionmodeling the PAS of a LOS indoor scenario is given in [139]. Reflections from distantscatterers located further around the transmitter and away from the receiver arrive primarily147from one direction with a narrow AS, while reflections of the local scatterers give rise to alarge AS. This phenomenon results in a Laplacian distribution with high occupancies at thecentral angle and lower ones at the larger angles [139]. In tunnels, on the other hand, due tothe presence of walls between the transmitter and the receiver, there is more than one strongray about 0o, the direct path between the transmitter and the receiver, as well as reflectionsfrom the sidewalls, which can give rise to a Gaussian distribution. Depending on the locationof the receiver respect to the transmitter (or tunnel zone), the standard deviation of thecorresponding Gaussian distribution is different. In the near zone, AS is large because ofstrong reflections coming from various angles, while in the far zone, it is small because ofgrazing incidence phenomenon, causing rays to arrive with angles very close to that of thedirect path, i.e., 0o.6.4.4 Antenna Interelement SpacingPearson?s correlation coefficient can be obtained from (4). In Figure 6-10, correlationcoefficient of the ULA antenna elements for different zones of two tunnels has beenobtained. For each zone, the zero-mean Gaussian fit of the corresponding PAS (with differentstandard deviations) has been considered. The results are compared for two tunnel transversesizes (5.1 m?3.8 m and 8 m ?5.6 m). As it can be seen, the largest difference between thetwo tunnels in terms of correlation coefficient is in the near zone, where higher order modes,which are dependent on the tunnel transverse size, are active and significantly impact the AS.On the other hand, in the far zone, where only lower order modes are active, for the sameantenna interelement separation, the correlation coefficient in both tunnels is very similar.For each zone, different interelement separation can be determined as the PAS foreach is different. From Figure 6-10, it is clear that ?/2 separation, which is a common spacing148for IEEE-802.11n-based off-the-shelf products, does not provide decorrelated MIMOsubchannels for most parts of both tunnels, and therefore, may not offer desirable capacity(close to that of an i.i.d. Rayleigh fading channel). This fact will be confirmed in thefollowing sections. By increasing the separation, lower correlation coefficient can beachieved. Capacity results presented in the next section confirm our results here. As we willsee by applying 2? separation, the mean capacity remains constant for the zone less than150 m and starts to decrease afterwards. However, by applying larger separations (e.g., 4?and 6?), the mean capacity remains constant for all the zones (10 m- 500 m).0 1 2 3 4 5 6 700. Coefficientd=10m-50md=50m-150md=150m-500m0 1 2 3 4 5 6 700. coefficient for different zones at 2.4GHzLarge tunnelSmall tunnelFigure 6-10 Correlation coefficient of ULA antennas vs. their separation for small and large tunnelsacross 10 m-500 m distance (multimode waveguide model).In [140], an expression is derived, which relates the antenna spacing of a ULA totheir correlation coefficient in an environment with Gaussian PAS. For a planar wave thatarrives from angle ? with respect to the normal to the array axis, assuming a Gaussian PAS149and ?? << 25o, the required separation to ensure correlation of |?| can be approximated by[140]:? ?? ??? ????? cos2ln2??d (8)where d, |?|, ?? and ?? are antenna separation, envelope correlation between two successiveantenna elements on the array, mean AOA (radians) and AS (radians), respectively. Becausewe found the PAS in tunnels follows a Gaussian distribution with angular spreads much lessthan 25o for most parts, (8) can also be used to calculate ULA antenna spacing. For a givenantenna separation, correlation coefficient obtained by formula (8) matches the one obtainedfrom Figure 6- Characterization of MIMO Channel Capacity6.4.5.1 Simulation SetupTo characterize the channel capacity, 4?4-MIMO matrices (H) were constructedacross the tunnel, H-matrices are normalized to their Frobenius norm, the operatingfrequency is considered to be 2.4 GHz, and the SNR is assumed to be fixed and equal to 20dB. To find the 4?4-MIMO capacity, the transmitting and receiving ULAs are placedhorizontally at the middle height (y=0). For the transmitter, a 4-element array is considered,which was placed perpendicular to the tunnel axis, and in the middle of the tunnel. For thereceiver, 4-element ULAs whose centres were located in an area of ????????4:2:4wwx ? ,? ?mmmz 500:1:10? were considered. The orientations were the same for the receiver and thetransmitter ULA. Intuitively, the chosen orientation for the ULA will provide optimumperformance because it occupies the largest space orthogonal to the arriving signals. In the150assessment of capacity, complementary-cumulative-distribution-functions (CCDF) of all thecapacities calculated in this area are considered, while for the mean capacity over distance, ateach longitudinal distance (z-axis) capacity was averaged over the tunnel width (x-axis). Mean Channel Capacity vs. DistanceIn this section, we evaluate 4?4-MIMO capacity vs. distance for several antennaspacings at SNR=20dB. Figure 6-11 compares capacity as a function of transmitter-receiverdistance for four interelement separations based on the multimode model. As it can be seen,for ?/2 interelement separation, capacity reduces with a sharp slope for very close distances(10 m-20 m). As Tx-Rx distance increases, the reduction rate gradually becomes less, andfinally becomes almost zero. The same behaviour was observed for the AS of the smalltunnel in Figure 6-8, which confirms the existence of high correlation between the channelcapacity and the AS of the surrounding environment. For 2? separation, mean capacitydecreases with a lower slope and for larger separations (4? and 6?) the reduction ratebecomes zero, and accordingly more consistent capacity can be achieved. These resultsreveal that ?/2 interelement separation shows lower capacity compared to the largerseparations for most locations inside the tunnel, however, for very close distances (10 m-20 m), due to sufficiently large AS, ?/2 separation also shows high capacity.15110 50 100 150 200 250 300 350 400 450 5008101214161820222426Capacity of the small tunnel for several antenna spacings at 2.4 GHzDistance (m)Capacity (b/sec/Hz)6?4?2??/2Figure 6-11 Comparison of 4?4-MIMO capacity vs. Tx-Rx distance for 4 antenna spacings(SNR=20 dB). MIMO Capacity CCDFIn this section, we find 4?4-MIMO capacity CCDFs at different tunnel zones and forseveral antenna spacings at SNR = 20dB. This can help to assess and quantify the medianand outage capacity improvement by increasing the interelement separation at differenttunnel zones: (1) 10 m-50 m, (2) 50 m-150 m, and (3) 150 m-500 m. The results aresummarized in Table 6-3. Based on these results, we can quantify the improvement ofmedian and %10 outage capacity achieved by increasing the interelement separation from ?/2to 2?, 4?, and 6?.As can be seen, significant improvement in capacity statistics can be observed byincreasing the separation from ?/2 to 2?. By increasing the spacing to 6?, furtherimprovement can be achieved, and capacity statistics becomes constant over the whole tunnelregardless of region. As a result, the increase in antenna spacing not only increases themedian and 10% outage capacity, but also increases the performance (capacity) consistency152for different longitudinal distances. To have a better comparison, Figure 6-12 is presented,which shows capacity CCDFs for two interelement separations of ?/2 and 6? in differenttunnel zones. In the near zone, capacity is higher due to the larger AS, resultant from largernumber of active propagation modes compared to the far zone. Therefore, in 10 m-50 mregion, CCDFs of both spacings (?/2 and 6?) are closer than those of other zones, meaningthat the capacity in this area is less affected by the array element spacing.Considering the CCDFs given in Figure 6-12 and Table 6-3, for the entire region(10 m-500 m), we can see that for ?/2 antenna separation (off-the-shelf products) the %10outage capacity is 9 bit/sec/Hz, while for 6?, it is 20.7 bit/sec/Hz. Therefore, 11.7 bit/sec/Hzimprovement on %10 outage capacity can be achieved by increasing separation to 6?.Similarly, improvement of the median capacity is found to be 12.6 bit/sec/Hz. Additionally,difference of outage capacities, and also median capacities in different zones becomeinsignificant (more consistency across the tunnel).8 10 12 14 16 18 20 22 24 26 2800. CCDF of a small tunnel for two antenna spacings at 2.4 GHzPr(Abscissa > C)ChannelCapacity (bit/sec/Hz)10m-50m50m-150m150m-500m10m-500m?/26?Figure 6-12 4?4-MIMO capacity CCDFs for several zones (10 m- 500 m) and two interelementseparations of ?/2and 6? (SNR=20 dB).153Table 6-3 Median and 10% outage 4?4 MIMO capacity (bit/sec/Hz) at SNR=20 dB for severalinterelement spacings and different zones of the small tunnel.Distance (m) 10 -50 50-150 150-500 10-500?/2 Cout 14.7 10.5 8.9 8.9Cmedian 19 12.8 10.4 112? Cout 17.4 19.2 13.3 14Cmedian 22.8 22.8 18.4 19.34?Cout 17 21 18.8 19Cmedian 21 23.7 22.8 22.96? Cout 18 20.4 21 20.7Cmedian 22.5 23.3 23.7 23.6i.i.d. RayleighCout 21 21 21 21Cmedian 24 24 24 246.5 Extension of the IEEE 802.11n MIMO Channel Model toUnderground MinesThe MIMO channel model developed by the IEEE 802.11n channel modelingcommittee can be extended to underground mines using the results presented here togetherwith a model for the channel impulse response. Using the PAS shape, AS, mean angle-of-arrival (AoA), and individual tap powers, correlation matrices of each tap can be determinedas described in [132]. For the uniform linear array (ULA), the complex correlationcoefficient at the linear antenna array is expressed as:r= RXX D( ) + jRXY D( ) (9)where D = 2 pdl and RXX and RXY are the cross-correlation functions between the real parts(equal to the cross-correlation function between the imaginary parts) and between the realpart and imaginary part, respectively, with:154RXX D( ) = cos Dsinf( )PAS f( ) df?pp?RXY D( ) = sin Dsinf( )PAS f( ) df?pp????????(10)6.6 ConclusionsThis chapter has focused on theoretical MIMO channel characterization, withparticular interest on characterization of angular properties in underground mine tunnels.Angular properties, correlation of array elements and channel capacity for different tunnelzones have been studied based on the multimode waveguide model.Based on our study, we conclude that the PAS inside a tunnel can be modeled by azero-mean truncated-Gaussian function with an azimuth AS between 4.5o and 23o, dependingon the Tx-Rx distance and the transverse dimension of the tunnel. For the near zone, the ASis large and for the far zone, it is small. For different zones and also for the entire tunnel, wefound that the AS found is slightly larger in tunnels with larger transverse dimension.However, as Tx-Rx distance increases, the difference between AS of different size tunnelsbecomes insignificant, and therefore, it becomes independent of tunnel size.For a typical mine size, multimode waveguide model predicts that AS starts at about24o at close distances to the transmitter (10 m), decreases with a sharp slope, and remainsconstant equal to 4o at about 200 m away from the transmitter. Small AS for most parts of themine, which is much smaller than AS of indoor environments (12o- 40o) requires customizedMIMO antenna configuration design for underground tunnels. Our study shows that bycareful antenna design and deployments, small AS (channel impairment for MIMO systems)155can be overcome, and mining industry can benefit from MIMO technology for undergroundcommunications.Several spacings (2?, 4?, and 6?) at the transmitter and the receiver were chosen forperformance assessment and comparison. Based on the correlation and capacity analysis, wefound that ?/2 separation offers poor performance for a 4?4-MIMO system in most parts ofthe tunnel, and therefore, the element spacings of the array should substantially be increasedto achieve desirable performance everywhere inside the tunnel, including at large Tx-Rxdistances (more than 300 m). The choice of antenna separation for the ULA can differdepending on the transverse dimension and the coverage area inside the tunnel.By increasing the interelement separation, capacity CCDFs show smaller standarddeviation (less steep). This implies that larger separations not only offer higher median andoutage capacity but also higher reliability (smaller standard deviation). Additionally, byincreasing the interelement separation to 6?, CCDFs of different zones become consistentover a range of 500 m, meaning that capacity performance does not degrade as distanceincreases, and therefore, performance consistency has also been achieved.Zone-specific characterization proposed in this study, can be used in undergroundtunnels, and particularly in mines, in which depending on the application, coverage zonesmay differ from mine to mine and even from level to level within the same mine. Suchstudies can accelerate deployment of MIMO-based systems for revolutionary applicationssuch as machine-to-machine communications, which requires highly reliable and efficientcommunications platform.156CHAPTER 7: OPTIMIZATION OF ANTENNAPLACEMENT IN DISTRIBUTED MIMOSYSTEMS FOR UNDERGROUND MINES7.1 IntroductionMIMO-based systems promise an increase in capacity only if the fading signalsobserved at the receiving antenna elements are decorrelated. If the angular spread of thesignals that arrive at the receiving antenna is low and the spacing of the antenna elements isinsufficient, MIMO capacity will be degraded. Because environments such as tunnels andmines show a smaller angular spread than conventional environments do, they are subject tomore of this degradation, and therefore careful consideration is required for the antennadesign and deployment.One of the most recent approaches for MIMO capacity enhancement is employingdistributed antennas across the area of coverage [141]-[145]. Distributed antenna systems(DAS) were originally proposed to extend coverage and to decrease delay spread [146] inconventional indoor environments, mines and tunnels. However, interest in DAS has beenrecently renewed due to its ability to provide high quality-of-service, universal coverage andhigh capacity. For such deployments, independent antenna nodes separated spatially,cooperate as a single MIMO system. Therefore, the role of DAS has evolved from merelybeing a repeater to being a sophisticated but a cost-effective solution that only offers bothextended coverage and increased capacity.157Extensive Monte-Carlo-based studies and modeling of DAS (or D-MIMO, i.e.,Distribured MIMO) have been performed in recent years with the main focus on theinformation theory and space-time coding aspects. Most of them have been developed basedon a generalized model for distributed antenna that was proposed in [141], [142]. This modelassumes independent small scale fading and uncorrelated shadowing, and exploits small scalefading and shadow fading simultaneously. In [147], the authors provided a closed-formexpression for the cell averaged ergodic capacity of D-MIMO, however, their modeling isbased on the assumption of high signal-to-noise-ratio (SNR) and uniform user distribution.More complex models have also been developed by relaxing the assumptions of high SNR,considering correlation of small scale fading with lognormal large scale fading andconsidering non-uniform mobile user distributions [142]-[144].Based on the aforementioned modelings and assumptions, general insights have beenproposed by a few studies in locating base stations (BS) in outdoor environments whileoptimizing capacity [144],[145]. Most of these studies have evaluated capacity enhancementachieved by employing D-MIMO in comparison with C-MIMO, i.e., Conventional MIMO orColocated MIMO. They have also addressed common challenges such as power imbalanceamong distributed antennas in D-MIMO systems, which is due to the large separationbetween them. Because most of these theoretical works are based on Monte-Carlosimulations and statistical channel models that are developed for conventional indoor andoutdoor environments, they may not be applicable to underground tunnel environmentsbecause of their different propagation behavior. DAS has been considered in transportationtunnels previously for coverage enhancement only [73],[148]. However, no previous study158concerning capacity enhancement of D-MIMO systems in underground tunnels and mineshas previously been conducted.For our analysis of D-MIMO for access-point (AP) to mobile communications, wehave used a deterministic model based on the multimode waveguide model [74], which hasrecently been developed to model electromagnetic (EM) propagation in underground minesand tunnels. Although this theoretical model has been experimentally validated forunderground environments [74], we have also verified it experimentally to ensure theaccuracy of our findings for short underground mine tunnels.By employing this model to find the channel-gain-matrix (H-matrix), we haveoptimized and compared the capacities and power distributions of C-MIMO and D-MIMOsystems by finding optimal locations for the APs. While the main focus of our performanceassessment in this study is on received power and channel capacity, we have also studiedanother performance measure, i.e., the condition-number (CN) of the H-matrix, in some partsof our analysis.For the optimization part, we have chosen the particle-swarm-optimization (PSO)method developed almost twenty years ago by a social psychologist and an electricalengineer [149]. This method is a widely used optimization method and well known withinthe EM community for its successful use in antenna design [150],[151]. The PSO is a globaloptimization method that performs guided random search technique over discrete searchspaces. The strengths of the PSO algorithm, including its robustness in overcoming the localminima problem, its ease of implementation, and its relatively fast convergence, make itsuitable for our application.159The remainder of this chapter is organized as follows. In Sec. 2, we describe thetheoretical basis of the multimode waveguide model. In Sec. 3, we explain our methodology,including details of the PSO method, the relevant performance measures, our modelassumptions and scenarios. In Sec. 4, we discuss our results and significant findings. Finally,in Sec. 5, we summarize the main outcomes of this study.7.2 Multimode Waveguide Modeling and Experimental ValidationMultimode modeling, which was proposed in [74], can be used to characterizewireless propagation in tunnels. It is an accurate theoretical model for high frequencies(upper ultra-high-frequency band and above) that takes into account higher order propagationmodes. By considering higher order modes as well as the dominant mode, this modelaccurately characterizes not only the far field but also the near field of an empty tunnel [74],and therefore can be expected to be applicable to short underground mine drifts and tunnelsas well. Here, we have used this model to find the H-matrix and the channel capacity(b/sec/Hz). In the following sections, the formulas used for multimode modeling are given,then experimental validation of the model for a short mine tunnel is presented and discussed.7.2.1 Multimode Waveguide ModelIn the multimode model, the tunnel is considered to be an oversized dielectricwaveguide. Assuming the excitation is a y-directed (vertically polarized) dipole, the fielddistribution eigennmE , of y-polarized hybrid modes at any position (x, y, z) inside the tunnel canbe found by solving Maxwell?s equations for the specified cross-sectional (transversal)dimensions and the material of the tunnel in the form of the eigenfunctions [130],[74]:160? ? ?????? ??????? ?? yxeigennm yhnxwmyxE ???? sincos,,,(1)where w and h are the tunnel width and height, respectively, and ?x and ?y are paritydesignations which take values of 0 or ?/2 depending on whether the integer values of m andn are even or odd [130]. The intensity of each mode depends on the excitation that can befound by applying the geometrical-optical (GO) model in the excitation plane (transmittingplane), i.e., the cross-sectional plane that contains the transmitting antenna. The EM fielddistribution on the transmitting plane is, in fact, the sum of the fields of all modes. The modeintensities are estimated by a mode matching technique as given in Equation (2) [74] which,assumes that the transmitter is vertically polarized (y-polarized) and located at (x0, y0, z0):? ? ? ? ? ? 0,),(,,,1 100zzjeigennmm nmnRx mnmneyxEyxCzyxE ?????????? ?? (2)where Cmn, ?mn, and ?mn are the mode intensity in the excitation plane (which depends on thetransmitter location), the attenuation coefficient, and the phase shift coefficient, respectively.Using the above formulas, we have constructed 4?4 C-MIMO and D-MIMO channelmatrices (H-matrix) as follows:? ? ? ? jimnmn zzjm niieigennmjjjmnRxTxij eyxEyxCkGGh ??????????1 1,,),(2?? (3)where, i is the receiving antenna number, j is the transmitting antenna number, hij is thecomplex channel gain between the ith receiver and the jth transmitter, GTx and GRx are theantenna gains, k is the wave number, jmnC is the mode intensity of the jth transmitter and,( , )eigenm n i iE x y is thevalue of the (m,n) mode eigenfunction as observed at the ith receiver.1617.2.2 Experimental Validation of Multimode Waveguide ModelIn [74], the authors compared and validated the multimode model with experimentalwork conducted by another group [13] at the same frequency (450 MHz and 900 MHz) andover the same axial length (2500 m). The experimental work was conducted in an archedtunnel while the multimode modeling was applied for a rectangular tunnel with the samecross sectional area as the arched tunnel, i.e., 7.8 m ? 5.3 m. Their comparison shows thesame trend and a good match between the multimode model and the experiment, particularlyfor long distances between the transmitter and receiver. However, the verification for shortdistances is not clearly observable.Experimental verification of the multimode model for short distances is necessarybecause it has been claimed that the approach is also applicable to underground mine tunnels,which are much shorter than transportation tunnels. To validate this model for shortdistances, we used the data we collected for an ultra-wide-band-MIMO project in anunderground mine at Myra Falls in BC, Canada (Figure 7-1). The tunnel width and height are5.1 m and 3.8 m, respectively. The measurements were taken by moving the transmitter in2 m increments from 1 m to 49 m from the receiver for a total of 25 points, as shown inFigure 7-1. Due to the mine restrictions, it was not possible to go further than 49 m. For ourchannel measurements, we used an Anritsu MS2034A vector-network-analyzer (VNA) tomeasure frequency response over a span of 2.49-4 GHz with 551 frequency points, RF overfibre optic cable for range extension, and two biconical antennas. Additional detail regardingthe measurement setup can be found in Chapter 3.162Figure 7-1 Map of the Myra Falls mine in British Columbia, Canada and the transmitter (Tx) andreceiver (Rx) locations.Figure 7-2 compares our experimental results with multimode modeling at 2.49 GHzfor four scenarios, each of which is presented on a subplot. We have considered verticalpolarization for two different transmitter and receiver heights: close to the ceiling andmedium heights. Considering the fact that this mine, like most of mines has manyirregularities in its geometry, overall good agreement can be seen for all the cases and forsuch short distances. After 35 m, the measurements and the multimode model start to divergedue to the presence of a curvature in the drift, as shown in Figure 7-1. Curvature adds extraloss that is not captured by the multimode model [109].7.3 Optimization of C-MIMO and D-MIMO ConfigurationsIn this section, we elaborate on how combining the multimode model with a globaloptimization method optimizes the performance measures that are relevant to C-MIMO andD-MIMO systems. We consider several scenarios for each system and use performancemeasures to evaluate and compare their performance.1637.3.1 Particle Swarm Optimization MethodThe PSO algorithm optimizes a problem by iteratively trying to improve a candidatesolution with regard to a given measure of quality. We used this optimization algorithm tooptimize the average capacity inside the tunnel. We formulated the cost function as follows:cost =average(Capacity)desired(Capacity)??????2(4)where the average capacity is found by considering a grid of receivers, and desired capacityis the capacity of an independent-identically-distributed (i.i.d.) Rayleigh fading channel at agiven SNR.0 10 20 30 40 50-50-40-30-20-100Vertical Polarization (Middle Height)0 10 20 30 40 50-40-30-20-100Vertical Polarization (Ceiling)MeasurementMultimodeOnset of Tunnel CurvatureFigure 7-2 Experimental validation of multimode model in a short underground mine tunnel;antennas are placed at medium height and close to the ceiling.In the PSO method, the particles (APs) are initially placed randomly in a searchspace, and then move through the search space based on (5). Assume there are M particles in164a swarm, along with N geometrical parameters to be optimized. In our study, 10 AP sets areconsidered as the particles (M=10) and each AP set consists of 4 antennas, each of which hasa 3D location vector (x, y, z) (N=12). At each iteration, the particles? positions and velocitiesare updated and stored in X and V matrices, respectively. As (5) shows, at each iteration, thevelocity of each particle is determined by the distances from its current position to the?personal best? (pbest) and the ?global best? (gbest) locations. The pbest is each particle?spersonal knowledge, the locations where each particle (AP) attains its best fitness value(average capacity) up to the current iteration, and gbest is the global (or swarm) knowledge,which is chosen from the M pbest?s, the location where the best fitness value was attained byany particle (locations of an AP set which gives the highest average capacity).? ? ? ?tttttttttVXXXGcXPcwVV?????????????1112211111 ?? (5)In (5), t denotes the present iteration, and the are M?N matrix P is the particles? bestlocal position, matrix G is the particles? best global position, matrix X is the particles?position and matrix V is the particles? velocity. Parameters c1 and c2 are scaling factors forbest personal and global positions, and ?1, ?2 are randomly chosen numbers uniformlydistributed over the interval [0,1]. Previous optimization studies recommend c1 and c2 to be1, however, they can be chosen based on numerical experiments. Inertia weight, w, is in therange of [0,1]. Numerical experiments indicate that the PSO algorithm converges faster if wis linearly damped with iterations starting at 0.9 and decreasing linearly to 0.4 at the lastiteration [152].1657.3.2 Two Performance Measures: Capacity and Condition NumberAccording to the literature, several performance measures for MIMO-based systemshave been considered, including channel capacity and the CN of the H-matrix [153]-[155].Channel capacity is defined as the highest transfer rate of information with arbitrary lowprobability of error, whereas CN is the ratio of the maximum to minimum singular values ofthe H-matrix with the desirable value of one. CN is often stated in dB form:)(log20))((log20)(1)(minmax1010minmax??????????HHdBCN(6)where ?, H, ?max, ?min, and CN are the linear condition number, channel gain matrix,maximum singular value, minimum singular value, and condition-number in dB,respectively.Although channel capacity and CN are associated in some ways, they are not fullycorrelated, and therefore CN cannot be a strong metric for capacity. Equations (6) and (7) [6]show CN and capacity, respectively. As can be seen, CN only considers the largest andsmallest singular values while the capacity depends on all singular values and theeigenstructure of the subchannels [156]:?? ???????? ?????????????????? ??kiiTxTx NSNRNSNRC1222 1logdetlog ?*N HHI Rx (7)where C, I, SNR, NRx, NTx, k, and ?i are capacity, identity matrix, average SNR, number ofreceivers, number of transmitters, number of singular values, and ith singular value,respectively.166In long-term-evolution (LTE) MIMO-based systems, CN is deterministicallycalculated from the instantaneous H-matrix (without the need for stochastic averaging) and isused to estimate the increase in SNR required for successful signal demodulation [157]. Itcan be directly measured by some of the recently developed vector signal analyzers[157],[158]. If CN is not too high, an increase in SNR may help to compensate the degradingeffect of CN. If it is too high, it implies that the H-matrix is ill-conditioned, and the systemwill suffer from irreducible error rates. Consequently, the system may need to change themultiplexing mode to other available modes, e.g., beamforming or diversity.7.3.3 Model Assumptions and ScenariosOur analysis focuses on AP-to-mobile communications in an empty tunnel, in whichthe mobile terminal is a laptop with four antennas positioned around the laptop screen asillustrated in Figure 7-3. The laptop screen is assumed to be 3? by 2?. The length of thetunnel is 100 m and the frequency of operation is assumed to be 2.4 GHz. Transmitter andreceiver antennas are vertically polarized with gain equal to one.Figure 7-3 D-MIMO and C-MIMO setups for AP-to-mobile communications inside an emptyrectangular tunnel.167For the C-MIMO case, four AP vertically polarized antennas are co-located in fourconfigurations: three uniform linear arrays (vertical, horizontal, and diagonal) and one squarearray. The horizontal array is along the tunnel width, close to the ceiling, the vertical array isalong the tunnel height and close to the sidewalls, the diagonal array is along the diagonal oftunnel cross-section, and the square configuration is aligned in parallel to the tunnel cross-section. The array along the tunnel axis was not considered because previous studies [21]show that it cannot offer significant capacity increase due to a high correlation betweenfading signals at the antenna elements. The linear arrays? interelement spacing is fixed andequal to 4? while the square array?s interelement spacing is 3? by 2?.For the D-MIMO case, the AP antennas are distributed along the tunnel. In all cases,300 iterations were performed to find the optimum locations for the AP antennas. In the threeC-MIMO cases, the optimum location for an AP with a fixed length array was sought, whilein the D-MIMO case, the locations for the distributed antennas were sought.  For the D-MIMO case, we considered three system sizes: (2?2), (3?3), and (4?4).Two scenarios of C-MIMO and D-MIMO systems for AP-to-mobile communicationsin a rectangular tunnel are given in Figure 7-3. In addition to vertical array, horizontal anddiagonal arrays have also been considered for the C-MIMO case, which are not depicted inFigure 7-3. Because this work focuses on propagation and channel modeling aspects,capacity has been optimized without relying on power control from the transmitter.Therefore, for capacity analysis of both C-MIMO and D-MIMO, we assumed that thechannel is known only at the receiver and equal power is radiated from each transmittingantenna. We set the PTx of each transmitting antenna to 0 dBm, and set the gains of theantennas at both the receiver and transmitter to one (GTx, GRx=1).168To find the AP positions that give optimum channel capacity, 300 iterations of thePSO algorithm were performed for mobile locations with the same height of 1 m over a gridof 1200 equally spaced points across the tunnel. For the D-MIMO case, four 3D positionvectors indicating the location of the AP antennas were found that maximize the averagecapacity over the grid of receivers. For the C-MIMO case, because the AP antennas are co-located, one 3D position for the whole array has been found. The PSO algorithm was runseveral times for each scenario to ensure the consistency of the results. Different parametersand their values for the multimode waveguide model and PSO method are presented inTable 7-1.7.4 Comparison of C-MIMO and D-MIMO System PerformanceIn this section, D-MIMO and C-MIMO performance are assessed and compared forseveral scenarios, including different system sizes, antenna configurations, antennaplacements, and tunnel lengths. Our objectives are to determine: 1) how much enhancementis achievable by employing better performing 4?4 C-MIMO antenna configurations, 2) howthe MIMO performance can be improved by optimally distributing the MIMO antennasacross the tunnel rather than placing them in the same tunnel transverse plane, and 3) how thetunnel length and antenna placement can impact D-MIMO performance.To accomplish these goals, we have considered different performance measures,including power distribution. In order to separately compare the power and otherperformance measures, first we analyze the power distribution, and then we study the CN andchannel capacity while using normalized H-matrices. We also elaborate on placementstrategies of AP antennas across the tunnel to enhance the capacity.169Table 7-1 Multimode and PSO simulation setup(w: tunnel width, h: tunnel height, l: tunnel length)MultimodeParametersTunnel dimension(w?h?l) 5.1?3.8?100 m3Tunnel materialcharacteristics? = 0.01? =5Antenna polarization VerticalFrequency of operation 2.4 GHzMobile height 1 mAntenna spacing atMobile: 3?,2?C-MIMO:AP-linear:4?AP-square:3?,2?PSOParametersNo. of iterations 300M: No. of particles 10N: No. of particles?parameters 12c1=c2 17.4.1 Power AnalysisFor the C-MIMO case, power analysis is performed for different antennaconfigurations whereas for the D-MIMO case, it is performed for different numbers of APs.In both cases, the transmit power of each antenna is assumed to be 0 dBm (PTx = 0 dBm), andto have a fair comparison between different cases with differing number of antennas, thenormalized received power of each case is plotted (Figure 7-4) using following formula:RxTxNormalized NNP2FH? (8)where ||H||F is the Frobenius norm of the H-matrix, and NTx, NRx are the number oftransmitters and receivers, respectively. In the case of a single antenna AP (1?1 system), theH-matrix only has one element. Figure 7-4 illustrates the power cumulative-distribution-functions (CDF) for different C-MIMO configurations. Among them, the diagonal170configuration shows better performance both in terms of %10 outage power (higher outagepower) and power variation across the grid of 1200 receivers (steepest CDF).-60 -55 -50 -45 -40 -35 -30 -25 -2000. Power (dBm)CDFC-MIMO Received Power for Different 4x4 AP ConfigurationsHorizontal ConfigurationVertical ConfigurationDiagonal ConfigurationSquare ConfigurationFigure 7-4 Received power CDFs for different C-MIMO access-point configurations.Because antennas are spatially distributed and far apart from each other in D-MIMOsystems, they are more subject to power imbalance, which is a degrading factor for D-MIMOsystems. However, due to the waveguide effect in tunnels, power loss along the tunnel axis islower compared to conventional environments, which makes D-MIMO systems even moresuitable for use in tunnels. Figure 7-5 shows power CDF plots for different size D-MIMOsystems where antennas are uniformly distributed inside the tunnel. In the case of a singleantenna AP, the antenna is located in the middle of the tunnel. As can be seen in Figure 7-5,the CDF of 4?4 D-MIMO is the steepest CDF, which implies that it has the least powervariation across the tunnel, and therefore is less vulnerable to power imbalance whereas 1?1has the highest variation (least steep CDF). Figure 7-5 shows how power distribution changesfor different size D-MIMO systems in a 100-m tunnel. Comparing the 4?4 D-MIMO case inFigure 7-5 with the cases in Figure 7-4, we can see that 4?4 D-MIMO offers lower powervariation across the tunnel compared to C-MIMO configurations with similar size.171-70 -65 -60 -55 -50 -45 -40 -35 -30 -2500. Power (dBm)CDFD-MIMO Power for Different System Sizes (Ptx=0 dBm)4x42x23x3SISO:1x1Figure 7-5 Received power CDFs for different D-MIMO system sizes.The aforementioned setup for APs in which they were equally and uniformlydistributed along the tunnel offers relatively balanced power, however, it does not necessarilyoffer optimum capacity and CN.  In the next section, we will see that performing the PSOalgorithm to distribute the antennas optimally can optimize capacity.7.4.2 Capacity and Condition Number AnalysisFor capacity analysis of C-MIMO, we have used (1) - (3) to construct 4?4 C-MIMOchannel gain matrices and calculate channel capacity [6]:???????????????? ?? *N HHI RxTxNSNRC detlog 2 (9)where ? denotes the transpose-conjugate, NTx is the number of the transmitting antennas, NRxis the number of receiving antennas, H is the NRx?NTx normalized channel gain matrix (basedon the Frobenius norm) and we have assumed that the NTx sources have equal power and areuncorrelated. Because the H-matrix is normalized in this formula, pathloss is removed, andtherefore capacity is mainly controlled by the degree of correlation between the subchannels.172Figure 7-6 shows the capacity CDFs of different 4?4 C-MIMO configurations. Theyare all lower than the capacity for the case with i.i.d. Rayleigh fading channels. As can beseen, C-MIMO capacity is highly dependent on the AP configurations. MIMO horizontal andvertical array configurations offer very low capacity in most parts of the tunnel. The diagonalconfiguration offers the best performance in terms of both power and capacity, but may notbe a practical solution. Square C-MIMO shows similar average capacity to diagonalconfiguration, however, its outage capacity is much lower than that of diagonal and itscapacity variation across the tunnel is very large compared to its counterparts.12 14 16 18 20 22 24 2600. (b/sec/Hz)CDFC-MIMO Capacity for Different 4x4 AP ConfigurationsHor. Config.Vert. Config.Diag. Config.Sq. Config.i.i.d. Rayleighi.i.d. RayleighFigure 7-6 Capacity CDFs for different C-MIMO access-point configurations.We increased the interelement spacing between the AP antennas for C-MIMO casesbut no improvement was achieved. This may be due to limited size of the array at the mobileterminal and the low angular spread inside the tunnel. For such scenarios, the angular spreadis too low to provide sufficient decorrelation among the C-MIMO subchannels, and due tothe constraints imposed by the size of the tunnel cross-section, further increasing the elementspacing is not possible. Figure 7-7 compares D-MIMO capacities with their i.i.d. Rayleighcounterparts. They are all close to i.i.d. Rayleigh capacity except for the 4?4 case. This173shows that increasing the number of antennas increases the capacity, however, fullmultiplexing gain may not be achieved.5 10 15 20 25 3000. Capacity (b/sec/Hz)CDFCapacity of Different Size D-MIMO SystemsD-MIMO: 2x2D-MIMO: 3x3D-MIMO: 4x4Rayleigh: 2x2Rayleigh: 3x3Rayleigh: 4x4Rayleigh:2x2 Rayleigh:3x3 Rayleigh:4x4Figure 7-7 Capacity CDFs for different D-MIMO system sizes.The CNs of all C-MIMO configurations were found to be very poor for most cases,which confirms that full multiplexing gain cannot be achieved by the 4?4 C-MIMO systemswe have considered in the tunnel under study. Figure 7-8 compares CN of D-MIMO systems.As it is shown, 2?2 D-MIMO shows the best CN, which is very close to that of an i.i.d.Rayleigh fading channel. For larger size D-MIMO systems, higher capacity can be achieved(as is shown in Figure 7-7), however, they are more susceptible to noise (as is shown inFigure 7-8).From Figure 7-5, Figure 7-7, and Figure 7-8, it can be concluded that among thescenarios that we considered, 2?2 D-MIMO shows desirable performance in terms of power,capacity and CN. By employing 3?3 D-MIMO, high capacity is still achievable, however, itsCN shows it is more susceptible to the noise compared to 2?2 D-MIMO systems. For largersize D-MIMO systems, if full multiplexing gain is not achievable, it may be more appropriate174to consider both multiplexing and diversity gains to obtain enhancement of the capacity aswell as reliability.0 10 20 30 4000.51CDFD-MIMO Condition Number0 10 20 30 4000.51Condition Number (dB)CDF2x2 D-MIMORayleigh4x4 D-MIMORayleigh0 10 20 30 4000.51CDF3x3 D-MIMORayleighFigure 7-8 Condition number CDFs for different D-MIMO system sizes.7.4.3 Impact of Antenna Placement on 2?2 D-MIMO PerformanceIn this section, we show how much performance enhancement is achievable byproperly positioning APs in the cross-sectional plane of the tunnel. For this purpose, we haveconsidered the following cases: (1) APs are distributed uniformly along the tunnel axis(1 2100 2 100,3 3z z?? ? ) and placed at the center of the tunnel (x = 0, y = 0), (2) APs aredistributed uniformly along the tunnel axis, placed at the medium height (y = 0) and close tothe sidewalls (1/5th ? 1/4th of the tunnel width from the sidewalls), (3) APs are firstdistributed uniformly along the tunnel axis, and then PSO finds the cross-sectional (xy-plane)locations that optimize capacity (PSO algorithm with constraint), and (4) PSO finds APs?175locations everywhere inside the tunnel that optimize capacity (PSO algorithm with noconstraint). Final results are given in Figure 7-9.Figure 7-9 shows that although uniformly distributing the APs along the tunnel axisand locating them at the center of the tunnel offer optimum power distribution, optimumcapacity may not be achieved. On the other hand, AP locations that offer optimum capacitydo not necessarily lead to optimum power distribution. Two other cases that are very similarfall in between. By running PSO several times, we have observed that by first uniformlydistributing APs along the tunnel axis, and then placing them in 1/5th ? 1/4th of the tunnelwidth from the sidewalls with medium height (y = 0), desirable performance (both capacityand power distribution) can be achieved. In addition, we found that performance may beimproved further by shifting antennas from the medium height towards the ceiling (1/5th ?1/4th of the tunnel height from the ceiling).  Physical interpretation for achieving betterperformance from placing APs at off-centered locations in the transverse plane can beattributed to the excitation of larger number of waveguide propagation modes in off-centeredlocations, which leads to an increase in decorrelation of D-MIMO subchannels.Based on the aforementioned results, the following strategy can be employed for APantenna deployment based on D-MIMO system. First, based on desired power and capacity,the number of APs should be set. Then, APs should be uniformly distributed along the tunnelaxis (z-axis) and placed close to the sidewalls and ceiling. To find the optimal cross-sectionallocation, which may differ from mine to mine and for different size MIMO systems (e.g.,3?3 or 4?4), position adjusments (fine position tuning) can be performed on the cross-sectional plane. By uniformly distributing APs, one is less likely to encounter powerimbalance problems, and therefore one wll be less dependent on AP power control at the176transmitter. Furthermore, by placing APs in the appropriate cross-sectional position,desirable capacity is achievable without degrading the power (as shown in Figure 7-9).-65 -60 -55 -50 -45 -40 -35 -3000. (dBm)CDF2x2 D-MIMO for 100 m TunnelAPs at tunnel centerAPs close to sidewallsOpt. capacity- APs uniformly along z-axisOpt. capacity - no constraint7 8 9 10 11 12 13 1400. (b/sec/Hz)CDFFigure 7-9 2?2 D-MIMO system performance for several scenarios.7.4.4 Impact of Tunnel Length on D-MIMO PerformanceIn this section, we show how tunnel length impacts power and capacity distributionsof 2?2 D-MIMO systems. For this purpose, three tunnel lengths, 100 m, 300 m, and 600 m,are considered. Figure 7-10 and Figure 7-11 show power and capacity distributions of a 2?2D-MIMO configuration for different tunnel length and different placements of AP in thetransverse plane. The results show that for a 2?2 D-MIMO system, the power distribution ismore sensitive to the tunnel length while the capacity distribution is more sensitive to thecross-sectional location of APs. In addition, Figure 7-11 shows that in all cases, the capacityis higher for AP locations close to the sidewalls compared to the center location.177-65 -60 -55 -50 -45 -40 -35 -3000. (dBm)CDF2x2 D-MIMO Power for Different Tunnel LengthsDash Lines: AP's are Close to the SidewallSolid Lines: AP's are at Center of the TunnelTunnel Length=300mTunnel Length=600mTunnel Length=100mFigure 7-10 2?2 D-MIMO power distribution for different cross-sectional locations of APs in tunnelswith different lengths.7 8 9 10 11 12 13 1400. (b/sec/Hz)CDF2x2 D-MIMO Capacity for Different Cross-Sectional Locations100m300m600mAP's close to the SidewallAP's at the Center of the TunnelFigure 7-11 2?2 D-MIMO capacity distribution for different cross-sectional locations of APs intunnels with different lengths.For AP positions that offer higher capacity, i.e., close to the sidewall and ceiling(about 1/5th - 1/4th of the width and height), we have also compared and evaluated differentsize D-MIMO systems for tunnels with different lengths. As Figure 7-12 illustrates, 2?2 D-MIMO capacity remains almost the same for the three considered lengths, whereas the largerD-MIMO sizes; capacity decreases as the tunnel length increases. As it can be seen, the178deviation of capacity CDFs for different tunnel lengths increases as the size of D-MIMOsystem increases.8 10 12 14 16 18 20 22 2400. (b/sec/Hz)CDFCapacity of Different Size D-MIMO Systems3x3 D-MIMO2x2 D-MIMO 4x4 D-MIMOSolid Lines: 100mDash Lines: 300mDot: 600mFigure 7-12 Comparison of different size D-MIMO systems for several tunnel lengths with APlocations close to the sidewall and ceiling (1/5th- 1/4th of the width and height).7.5 ConclusionsIn this study, we have developed a novel technique to assess, compare and optimizethe performance of co-located and distributed MIMO systems for AP-to-mobilecommunications in underground short tunnels (e.g., mines). This technique combines therecently developed multimode waveguide model with the PSO global optimization method tofind optimal locations of APs in a short rectangular tunnel. Because the verification of themultimode waveguide model performed previously for short tunnels was not as clear as it isfor long ones, we verified it with our experimental work in an underground mine to ensurethe applicability of the model to short tunnels.Assessing several C-MIMO antenna deployments for AP-to-mobile communications,we found that C-MIMO may not show the performance it is capable of in above-ground179environments, which can be attributed to the very low angular spread in underground tunnelsand high correlation among antenna elements at mobile station. The common practice ofincreasing the spacing of array elements may not be considered as a solution because of theconstraints imposed by the tunnel cross-sectional size. This study reveals that low angularspread can be overcome by employing D-MIMO systems. In comparison with C-MIMOdeployments, significant improvement in capacity and power distribution can be achieved.However, the level of the improvement is critically dependent on the AP locations. If APantennas are optimally distributed across the tunnel, they can noticeably outperform co-located deployments.The results of this study have allowed us to develop a deployment strategy for D-MIMO based APs, which takes into account both channel capacity and power distributionacross the tunnel, and can be summarized as follows: 1) uniformly distributing the APs alongthe tunnel (for desirable power distribution) and 2) optimizing their locations on thetransverse plane (tuning for higher capacity); in this way higher capacity can be achievedwithout degrading the power distribution. The optimum locations of D-MIMO antennas arefound to be close to the sidewalls (about 1/5th-1/4th of the tunnel width from the wall) andceiling (about 1/5th-1/4th of the tunnel height from the ceiling), as in these locations largernumber of higher order waveguide modes can be excited. We also showed that tunnel lengthand antenna transverse location impact the power distribution and channel capacity ofdifferent size D-MIMO systems. For 2?2 D-MIMO systems in particular, the impact oftunnel length found to be more evident on the power distribution, while the impact of antennatransverse location found to be more evident on the capacity.180The results of this study confirm that D-MIMO systems can be a suitable alternativeto facilitate development of applications that require high data-rate in underground tunnelsand mines. One of such applications can be transmission of real-time 3D high definitionvideo streaming of the mining process from the face area to the surface or other safeterminals to enhance tele-mining procedure, which makes mines safer and more productivework places. This work can be improved further by including the impact of tunnel geometry(e.g., curvature) and infrastructure in the model that are expected to affect the performance ofD-MIMO systems.181CHAPTER 8: CONCLUSIONSWe have achieved our objectives and thereby contributed to the efficient and effectivedeployment of MIMO-based wireless systems in underground mines and transportationsystems.8.1 ContributionsThe overarching contributions of this work are: 1) determination of the reduction inthe angular spread of multipath signals that arrive at the receiver in an underground minecompared to that observed in conventional surface environments and the manner in which itdecreases with increasing transmitter-receiver separation and 2) demonstration that theantenna elements in MIMO antenna arrays used in underground environments must thereforebe separated by several wavelengths (rather than the customary half-wavelength used insurface environments) in order to achieve acceptable performance. Further, the separationbetween the antennas must increase as the transmitter-receiver separation increases, higherorder modes attenuate and, as a consequence, angular spread decreases.More specific contributions are as follows: Our MIMO channel measurement system(described in Chapter 3) performed well and allowed us to conduct measurement campaignsin relatively harsh underground mine and tunnel environments including a service tunnelunderneath the Woodward Instructional Resources Centre at the University of BritishColumbia and a lead-zinc mine operated by Nystar Mining at Myra Falls, BC. Since then, the182underlying platform has become a permanent part of our lab?s research infrastructure and hasbeen used by several other teams to conduct their own measurement campaigns.The measurement campaigns conducted in the service tunnel at UBC and theunderground mine at Myra Falls (described in Chapters 4 and 5) allowed us to verify that thetrends and values revealed by ray-tracing and multimode waveguide modelling aresubstantially correct despite the simplifications and assumptions inherent to theseapproaches. Among different practical scenarios, we identified two better performing antennaarray configurations: (1) a horizontal array oriented perpendicular to the tunnel axis andplaced under the ceiling with vertically polarized antennas, (2) a vertical array placed close toone of the sidewalls with horizontally polarized antennas. These two better performing arrayconfigurations perform differently depending on the existence of extensive infrastructure.The results obtained during our measurement campaigns also allowed us to verify that themultimode waveguide model accurately predicts path loss for transmitter-receiverseparations of less than 100 metres (as described in Chapter 6).The multimode waveguide model (described in Chapter 6) allowed us to characterizeangular-spread and MIMO channel capacity and reveal that the angle of arrival distributionin underground mines closely follows a Gaussian distribution and the manner in whichangular spread decreases with increasing transmitter-receiver separation. We note that theseobservations are sufficient to permit the correlation matrix required to extend the IEEE802.11n MIMO channel model to underground mining environments.Finally, we have used the multimode waveguide model in combination with particleswarm optimization (PSO) to realize a novel technique for assessing and optimizing theperformance of Distributed-MIMO (D-MIMO) antenna systems in underground mine183environments (as described in Chapter 7). While D-MIMO systems are deployed in otherenvironments due to mitigating shadowing (diversity gain) or multiplexing gain, shadowingis not a concern in straight sections of tunnels. In such cases, we recognize for the first timethat the D-MIMO solution is more attractive for longer-range applications due to thereduction of angular spread versus distance compared to conventional MIMO. D-MIMOsystem performance is greatly dependent on the axial distribution of the antennas as well astheir location in the cross-sectional plane.8.2 Recommendations for Future WorkThe focus of our work for the most part was on fundamental issues associated withwave propagation in straight sections. The next step will be to pursue a follow on study ofwave propagation in the vicinity of more complicated geometries such as bends andjunctions. Toward this end, it would be useful to establish the feasibility of combining themultimode waveguide model with ray-tracing so that runtime and implementation issues forhandling more complex and arbitrary geometries (e.g., walls? tilt, indentation, curvature,branches, and infrastructure) can be resolved. This can be implemented by dividing the studyarea into two sections separated by a planar boundary. One section can have arbitrarygeometry and the other one should have a canonical waveguide shape. EM field distributionon the boundary plane (excitation plane) can be determined using the ray-tracing method.This field distribution can be viewed as the weighted sum of the field of all modes. The modeintensities are estimated by a mode-matching technique. Once the mode intensity isdetermined in the excitation plane, the EM field in the rest of the tunnel can be predicted bysumming the EM field of each mode.184It would be useful to determine whether it is necessary to include surface roughness inmodelings (while it is desirable to keep models as compact as possible). Surface roughnessmay be neglected for cases in which antennas are not very close to the wall/ceiling. However,it may need to be considered if antennas are very close or attached to the walls. In addition,to make the multimode waveguide model more general, the radiation pattern of the antennacan also be included.185REFERENCES[1] ?Rock Solid: The Mining Industry in British Columbia 2008,?PriceWaterhouseCoopers LLP, Vancouver, BC, 5 May 2009, 32 pp.[2] P. Delogne, Leaky Feeders and Subsurface Radio Communication, Stevenage, UK:Peter Pegrinus, 1982.[3] P. Delogne, ?Mini-review: EM propagation in tunnels,? IEEE Trans. AntennasPropag., vol. 39, pp. 401-406, 1991.[4] W. C. Jakes, Microwave Mobile Communications, New York: McGraw-Hill,pp. 110, 1974, p. 110.[5] D. J. R. Martin, ?Leaky-feeder radio communication: a historical review,? in Proc.IEEE VTC ?84, 21-23 May 1984, pp. 25-30.[6] J. Foschini and M. Gans ?On the limit of wireless communications in a fadingenvironment when using multiple antennas,? Wireless Personal Commun., vol. 6,pp. 311-335, 1998.[7] P. Almers, et al., ?Survey of channel and radio propagation models for wirelessMIMO systems,? EURASIP J. Wireless Commun. Networking, vol. 2007, article ID19070, 19 pp.[8] P. E. Mogensen et al., ?Antenna arrays and space division multiple access,? in GSMEvolution Towards 3rd Generation Systems, Editors: Z. Zvonar, P. Jung and K.Kammerlander, Dordrecht, The Netherlands: Kluwer Academic Publishers, 1999,pp. 122-123.[9] K. Haneda, J. Poutanen, L. Liu, C. Oestges, F. Tufvesson, and P. Vainikainen,?Comparison of delay and angular spreads between channel measurements and theCOST2100 channel model,? in Proc. IEEE AP-URSI, Nov. 2010, pp. 477-480.[10] A. Paulraj, R. Nabar, and D. Gore, ?Space-Time Propagation,? Introduction toSpace Time Wireless Communications, Cambridge: University Press, 2003, p. 24.[11] A. Emslie, R. Lagace, and P. Strong, ?Theory of the propagation of UHF radiowaves in coal mine tunnels,? IEEE Trans. Antennas Propag., vol. 23, pp. 192-205,1975.[12] M. Lienard and P. Degauque, ?Natural wave propagation in mine environments,?IEEE Trans. Antennas Propag., vol. 48, pp.1326-1339, 2000.[13] D. G. Dudley, M. Lienard, S.F. Mahmoud, and P. Degauque, ?Wireless propagationin tunnels,? IEEE Antennas Propag. Mag., vol. 49, pp. 11-26, 2007.[14] S. F. Mahmoud and J. R. Wait, ?Geometrical optical approach for electromagneticwave propagation in rectangular mine tunnels,? Radio Sci., vol. 9, pp. 1147?1158,1974.186[15] Y. Hwang, Y. P. Zhang, and R. G. Kouyoumjian, ?Ray-optical prediction of radio-wave propagation characteristics in tunnel environments, Part 1: Theory,? IEEETrans. Antennas Propag., vol. 46, pp. 1328-1336, 1998.[16] D. Didascalou, T. M. Schafer, F. Weinmann, and W. Wiesbeck, ?Ray-densitynormalization for ray-optical wave propagation modeling in arbitrarily shapedtunnels,? IEEE Trans. Antennas Propag., vol. 48, pp. 1316-1325, 2000.[17] A. H. Wong, M. J. Neve, and K. W. Sowerby, ?Antenna selection and deploymentstrategies for indoor wireless communication systems,? IET Commun., vol.4, pp.732-738, 2007.[18] C. Hermosilla, R. Feick, R. Valenzuela, and L. Ahumada, ?Improving MIMOcapacity with directive antennas for outdoor-indoor scenarios,? IEEE Trans.Wireless Commun., vol. 8, pp. 2177-2181, 2009.[19] P. Kyritsi, D. C. Cox, R. A. Valenzuela, and P. W. Wolniansky, ?Correlationanalysis based on MIMO channel measurements in an indoor environment,? IEEEJ. Sel. Areas Commun., vol. 21, pp. 713-720, 2003.[20] P. Kyritsi, D. C. Cox, R. A. Valenzuela, and P. W. Wolniansky, ?Effect of antennapolarization on the capacity of a multiple element system in an indoorenvironment,? IEEE J. Sel. Areas Commun., vol. 20, pp. 1227-1239, 2002.[21] M. Lienard, P. Degauque, J. Baudet, and D. Degardin, ?Investigation on MIMOchannels in subway tunnels,? IEEE J. Sel. Areas Commun., vol. 21, pp. 332-339,2003.[22] J. Molina-Garcia-Pardo, M. Lienard, P. Stefanut, and P. Degauque, ?Propagation intunnels: experimental investigations and channel modeling in a wide frequencyband for MIMO applications,? EURASIP J. Wireless Commun. Netw., vol. 2009,pp. 1-9, 2009.[23] M. Lienard and P. Degauque, ?MIMO communication in tunnels: Influence of therange and of the tunnel geometry on system performances,? Proc. WirelessCommun. Sys., 2004, pp. 290-293.[24] J. Molina-Garcia-Pardo, M. Lienard, P. Degauque, D.G. Dudley, and L. Juan-Llacer, ?Interpretation of MIMO channel characteristics in rectangular tunnels frommodal theory,? IEEE Trans. Veh. Technol., vol. 57, pp. 1974-1979, 2008.[25], visited on May 2012.[26] ?Mine 2010: Back to the Boom,? Price Waterhouse Coopers, 2010.[27] ?World mining equipment: Industry forecasts for 2013 and 2018,? FreedoniaGroup, 2009.[28] D. G. Large, L. Ball, and A. J. Farstad, ?Radio transmission to and fromunderground coal mines-Theory and experiment,? IEEE Trans. Commun., vol. 21,pp. 194?202, 1973.187[29] R. S. Nutter and M. D. Aldridge, ?Status of mine monitoring and communications,?IEEE Trans. Ind. Appl., vol. 24, pp. 820?826, 1988.[30] P. Delogne, ?The INIEX mine communications systems,? in Proc. Radio: RoadsTunnels Mines, Apr. 1974, pp. 129?136.[31] J. Murphy and H. E. Parkinson, ?Underground mine communications,? in Proc.IEEE, Jan. 1978, pp. 26?50.[32] D. J. R. Martin and R. Webster, ?The use of radio in British coal mines,? in Proc.Radio: Road Tunnels Mines, Apr. 1974, pp. 110?128.[33] D. J. R. Martin, ?Leaky-feeder radio communication: A historical review,? in Proc.IEEE VTC, May 1984, pp. 25?30.[34] P. Delogne, ?EM propagation in tunnels,? IEEE Trans. Antennas Propag., vol. 39,pp. 401?406, 1991.[35] ?Underground coal mining disasters and fatalities-U.S. 1900-2006.? [Online].Available:[36] L. K. Bandyopadhyay, S. K. Chaulya, P. K. Mishra, and A. Choure, ?Wirelessinformation and safety system for underground mines,? in Proc. URSI, Aug. 2008,pp. 9?16.[37] K. Srinivasan, M. Ndoh, and K. Kaluri, ?Advanced wireless networks forunderground mine communications,? in Proc. IEEE ICWCUCA, Jun. 2005, pp. 51?54.[38] W. H. Schiffbauer and J. F. Brune, ?Coal mine communications,? AmericanLongwall Mag., 2006.[39] P. Laliberte, ?Summary study of underground communications technologies-Finalproject report,? CANMET, 2009.[40] ?OMSHR.? [Online]. Available:[41] S. Yarkan, S. Guzelgoz, H. Arslan, and R. Murphy, ?Underground minecommunications: A survey,? IEEE Commun. Surv. Tutorials, vol. 11, pp. 125?142,2009.[42] T. Sarkar, Z. Ji, K. Kim, A. Medouri, and M. Salazar-Palma, ?A survey of variouspropagation models for mobile communication,? IEEE Trans. Antennas Propag.,vol. 45, pp. 51?82, 2003.[43] D. G. Michelson and S. S. Ghassemzadeh, ?Measurement and modeling of wirelesschannels,? in New Directions in Wireless Communications Research,, V. Tarokh,Ed. Springer, 2009, ch. 1, pp. 1?27.[44] W. Pittman and R. Church, ?Through-the-earth electromagnetic trapped minerlocation systems: A review,? Bureau of Mines, 1985.188[45] D. Hill and J. Wait, ?Theoretical noise and propagation models for through-the-earth communication,? U.S. Bureau Mines, 1982.[46] J. Durkin, ?Apparent earth conductivity over coal mines as estimated from through-the-earth electromagnetic transmission tests,? U.S. Bureau of Mines, 1984.[47] T. D. Barkand, N. W. Damiano, and W. A. Shumaker, ?Through-the earth, two-way, mine emergency, voice communication systems,? in Proc. IAS, Oct. 2006, pp.955?958.[48] C. J. Bise, R. Ferriter, J. Joy, J. A. Lamonica, D. Lauriski, J. Main, J. N. Murphy,and W. York-Feirn, ?Improving mine safety technology and training: EstablishingU.S. global leadership,? Mine Safety Technol. Training Commission, pp. 2?4, 2006.[49] ?Mine site technologies.? [Online]. Available:[50] A. Chehri, P. Fortier, and P. M. Tardif, ?UWB-based sensor networks forlocalization in mining environments,? Elsevier Ad Hoc Netw., vol. 7, pp. 987?1000,2009.[51] D. Frielos, ?Xstrata mines RFID?s benefits,? RFID J., 2007.[52] ?Becker Varis.? [Online]. Available:[53] D. A. Hill, Radio Wave Propagation Ground Effects, V. Tarokh, Ed. Wiley, 2005.[54] A. C. Tripp, R. McNearny, and C. Furse, ?Prof. James R. Wait and miningproduction technology-An appreciation,? IEEE Trans. Antennas Propag., vol. 48,pp. 1438?1441, 2000.[55] S. F. Mahmoud, ?On the attenuation of monofilar and bifilar modes in minetunnels,? IEEE Trans. Microw. Theory Techn., vol. 22, pp. 845? 847, 1974.[56] S. F. Mahmoud and J. R. Wait, ?Theory of wave propagation along a thin wireinside a rectangular waveguide,? Radio Sci., vol. 9, pp. 417?420, 1974.[57] D. T. Updyke, W. C. Muhler, and H. C. Turnage, ?An evaluation of leaky feedercommunication in underground mines,? U. S. Dept. Interior, Bureau of Mines,1980.[58] E. A. J. Marcatili and R. A. Schmeltzer, ?Hollow metallic and dielectric wave-guides for long distance optical transmission and lasers,? Bell Syst. Tech. J., vol. 43,pp. 1783?1809, 1964.[59] J. I. Glaser, ?Attenuation and guidance of modes on hollow dielectric waveguides,?IEEE Trans. Microw. Theory Techn., vol. 17, pp. 173?174, 1969.[60] C. Nerguizian, C. Despins, S. Affes, and M. Djadel, ?Radio-channelcharacterization of an underground mine at 2.4 GHz,? IEEE Trans. WirelessCommun., vol. 4, pp. 2441?2453, 2005.[61] S. R. Saunders and A. Aragon-Zavala, Antennas and Propagation for Wireless189Communication Systems. Chichester, England: Wiley, 2007.[62] J. R. Wait, ?A fundamental difficulty in the analysis of cylindrical waveguides withimpedance walls,? IEEE Electron. Lett., vol. 3, pp. 87?88, 1967.[63] J. M. Molina-Garcia-Pardo, M. Lienard, A. Nasr, and P. Degauque, ?On thepossibility of interpreting field variations and polarization in arched tunnels using amodel for propagation in rectangular or circular tunnels,? IEEE Trans. AntennasPropag., vol. 56, pp. 1206?1211, 2008.[64] J. C. Chiba, T. Inaba, Y. Kuwamoto, O. Banno, and R. Sato, ?Radio communicationin tunnels,? IEEE Trans. Microw. Theory Tech., vol. 26, pp. 439?443, 1978.[65] C. L. Holloway, D. A. Hill, R. A. Dalke, and G. A. Hufford, ?Radio wavepropagation characteristics in lossy circular waveguides such as tunnels, mineshafts and boreholes,? IEEE Trans. Antennas Propag., vol. 48, pp. 1354?1366,2000.[66] D. G. Dudley, ?Wireless propagation in circular tunnels,? IEEE Trans. AntennasPropag., vol. 53, pp. 435?441, 2005.[67] D. G. Dudley and S. F. Mahmoud, ?Linear source in a circular tunnel,? IEEE Trans.Antennas Propag., vol. 54, pp. 2034?2047, 2006.[68] Y. P. Zhang and Y. Hwang, ?Enhancement of rectangular tunnel waveguidemodel,? in Proc. IEEE APMC, Dec. 1997, pp. 197?200.[69] Y. P. Zhang, ?New model for propagation loss prediction in tunnels,? IEEE Trans.Veh. Technol., vol. 52, pp. 1308?1314, 2003.[70] Z. Sun and I. Akyildiz, ?Channel modeling and analysis for wireless networks inunderground mines and road tunnels,? IEEE Trans. Commun., vol. 58, pp. 1758?1768, 2010.[71] Y. P. Zhang, Y. Hwang, and R. G. Kouyoumjian, ?Ray-optical prediction of radio-wave propagation characteristics in tunnel environments-Part 2: Analysis andmeasurements,? IEEE Trans. Antennas Propag., vol. 46, pp. 1337?1345, 1998.[72] M. Lienard and P. Degauque, ?Propagation in wide tunnels at 2 GHz: A statisticalanalysis,? IEEE Trans. Antennas Propag., vol. 47, pp. 1322? 1328, 1998.[73] T. Klemenschits and E. Bonek, ?Radio coverage of road tunnels at 900 and 1800MHz by discrete antennas,? in Proc. IEEE PIMRC, 1994, pp. 411?415.[74] Y. P. Zhang, G. X. Zheng, and J. H. Sheng, ?Radio propagation at 900 MHz inunderground coal mines,? IEEE Trans. Antennas Propag., vol. 49, pp. 757?762,2001.[75] M. Boutin, A. Benzakour, C. L. Despins, and S. Affes, ?Radio wavecharacterization and modeling in underground mine tunnels,? IEEE Trans.Antennas Propag., vol. 56, pp. 540?549, 2008.[76] J. B. Keller, ?Geometrical theory of diffraction,? J. Opt. Sci. Amer., vol. 52,190pp. 116?130, 1962.[77] R. Kouyoumjian and P. Pathak, ?A uniform geometrical theory of diffraction for anedge in a perfectly conducting surface,? in Proc. IEEE, Nov. 1974, pp. 1448?1461.[78] S. F. Mahmoud and J. R. Wait, ?Geometrical optical approach for electromagneticwave propagation in rectangular mine tunnels,? Radio Sci., vol. 9, pp. 1147?1158,1974.[79] P. Beckmann and A. Spizzichino, The Scattering of Electromagnetic Waves fromRough Surfaces. Artech House, 1963, p. 124.[80] C. I. Beard, ?Coherent and incoherent scattering of microwaves from the ocean,?IRE Trans. Antennas Propag., vol. 9, pp. 470?483, 1961.[81] D. Didascalou, T. M. Schfer, F. Weinmann, and W. Wiesbeck, ?Ray densitynormalization for ray-optical wave propagation modeling in arbitrarily shapedtunnels,? IEEE Trans. Antennas Propag., vol. 48, pp. 1316?1325, 2000.[82] P. Mariage, M. Lienard, and P. Degauque, ?Theoretical and experimental approachof the propagation of high frequency waves in road tunnels,? IEEE Trans. AntennasPropag., vol. 42, pp. 75?81, 1994.[83] Y. Hwang, Y. P. Zhang, and R. G. Kouyoumjian, ?Ray-optical prediction of radiowave propagation characteristics in tunnel environments-Part 1: Theory,? IEEETrans. Antennas Propag., vol. 46, pp. 1328?1336, 1998.[84] L. Ramirez, F. Hasselmann, and Y. Zhang, ?Channel characteristics in tunnels:FDTD simulations and measurement,? J. Microw. Opt. Electromag. Appl., vol. 10,pp. 121?130, 2011.[85] M. F. Hadi and S. F. Mahmoud, ?Modeling wireless propagation in a rectangulartunnel with the compact-FDTD method,? in Proc. IEEE RWS, Jan. 2008, pp. 339?342.[86] Y. Wu, M. Lin, and I. J. Wassell, ?Modified 2D finite-difference time domain basedtunnel path loss prediction for wireless sensor network applications,? J. Commun.,vol. 4, pp. 214?223, 2009.[87] D. Didascalou, M. Dottling, N. Geng, and W.Wiesbeck, ?An approach to includestochastic rough surface scattering into deterministic ray-optical wave propagationmodeling,? IEEE Trans. Antennas Propag., vol. 51, pp. 1508?1515, 2003.[88] M. Ndoh, G. Y. Delisle, and R. Le, ?A novel approach to propagation prediction inconfined and diffracting rough surfaces,? Int. J. Numer. Model, vol. 16, pp. 535?555, 2003.[89] A. Chehri, P. Fortier, and P. M. Tardif, ?Characterization of the ultrawidebandchannel in confined environments with diffracting rough surfaces,? WirelessPersonal Commun., vol. 62, pp. 859?877, 2012.[90] S. Boutin, M.and Affes, C. Despins, and T. Denidni, ?Statistical modeling of a radio191propagation channel in an underground mine at 2.4 and 5.8 GHz,? in Proc. IEEEVTC, vol. 1, May-Jun. 2005, pp. 78?81.[91] Y. Rissafi, L. Talbi, and M. Ghaddar, ?Experimental characterization of an UWBpropagation channel in underground mines,? IEEE Trans. Antennas Propag., vol.60, pp. 240?246, 2012.[92] Y. P. Zhang and Y. Hwang, ?Characterization of UHF radio propagation channelsin tunnel environments for microcellular and personal communications,? IEEETrans. Veh. Technol., vol. 47, pp. 283?296, 1998.[93] J. S. Seybold, Introduction to RF Propagation. Wiley, 2005.[94] B. Nkakanou, G. Y. Delisle, and N. Hakem, ?Experimental characterization ofultra-wideband channel parameter measurements in an underground mine,?J. Comput. Netw. Commun., vol. 2011, pp. 1?7, 2011.[95] C. Nerguizian, C. Despins, and S. Affes, ?Geolocation in mines with an impulseresponse fingerprinting technique and neural networks,? IEEE Trans. WirelessCommun., vol. 5, pp. 603?611, 2006.[96] H. I. Volos, C. R. Anderson, W. C. Headley, R. M. Buehrer, C. R. C. Silva, and A.Nieto, ?Preliminary UWB propagation measurements in an underground limestonemine,? in Proc. IEEE GLOBECOM, Nov. 2007, pp. 3770?3774.[97] M. Lienard, P. Stefanut, and P. Degauque, ?Modeling and understanding MIMOpropagation in tunnels,? J. Commun., vol. 4, pp. 241?247, 2009.[98] Y. Sun and B. Zhang, ?System model of underground UWB based on MB-OFDM,?Int. J. Commun., Netw. Syst. Sci., vol. 4, pp. 59?64, 2011.[99] X. Yuan, ?Comb-type pilot-aided OFDM channel estimation for undergroundWLAN communications,? Ph.D. dissertation, Laval University, Canada, 2007.[100] H. Hashemi, D. Lee, and D. Ehman, ?Statistical modeling of the indoor radiopropagation channel: Part II,? in Proc. IEEE VTC, May 1992, pp. 839?843.[101] P. K. Misra, ?Underground mine location sensing system,? M.Sc. Thesis,University of New South Wales, Australia, 2009.[102] J. Karedal, S. Wyne, P. Almers, F. Tufvesson, and A. Molisch, ?A measurementbased statistical model for industrial ultra-wideband channels,? IEEE Trans.Wireless Commun., vol. 6, pp. 3028?3037, 2007.[103] D. Cassioli, M. Win, F. Vatalaro, and A. Molisch, ?Low complexity rake receiversin ultra-wideband channels,? IEEE Trans. Wireless Commun., vol. 6, pp. 1265?1275, 2007.[104] A. Chehri, P. Fortier, and P. Tardif, ?An investigation of UWB-based wirelessnetworks in industrial automation,? Int. J. Comput. Sci. Netw. Security, vol. 8,pp. 179?188, 2008.[105] L. Bernado, A. Roma, A. Paier, T. Zemen, N. Czink, J. Karedal, A. Thiel,192F. Tufvesson, A. F. Molisch, and C. F. Mecklenbrauker, ?In-tunnel vehicular radiochannel characterization,? in Proc. IEEE VTC, May 2011, pp. 1?5.[106] P. Delogne, Leaky Feeders and Subsurface Radio Communications. Peter Pegrinus,1982.[107] A. E. Goddard, ?Radio propagation measurements in coal mines at UHF and VHF,?in Proc. Through-Earth Electromagn., Aug. 1973, pp. 15?17.[108] Y. P. Zhang and Y. Hwang, ?Measurements and statistical modeling of 900 MHzradio propagation channels for microcellular and personal communications intunnels,? Wireless Personal Communic., vol. 7, pp. 25?39, 1998.[109] Y. Zhang, Y. Hwang, and P. Ching, ?Characterization of UHF radio propagationchannel in curved tunnels,? in Proc. IEEE PIMRC, Oct. 1996, pp. 798?802.[110] Y. P. Zhang, G. X. Zheng, and J. H. Sheng, ?Excitation of UHF radio waves intunnels,? Microw. Opt. Technol. Lett., vol. 22, pp. 408?410, 1999.[111] Y. P. Zhang, ?Enhancement of waveguide model for propagation-loss prediction intunnels,? Microw. Opt. Technol. Lett., vol. 30, pp. 10?12, 2001.[112] Y. Yamaguchi, T. Honda, and M. Sengoku, ?Reduction of wave propagation lossby mesh in rectangular tunnels,? IEEE Trans. Electromagn. Compat., vol. 37,pp. 88?93, 1995.[113] Y. P. Zhang, Y. Hwang, and R. Kouyoujian, ?Ray-optical prediction of radio-wavepropagation characteristics in tunnel environments, Part 2: Analysis andmeasurements,? IEEE Trans. Antennas Propag., vol. 46, pp.1337-1345, 1998.[114] Y. P. Zhang, ?Novel model for propagation loss prediction in tunnels,? IEEE Trans.Veh. Technol., pp. 1308-1314, 2003.[115] I. E. Telatar, ?Capacity of multi-antenna Gaussian channels,? Eur. Trans.Telecommun., vol 10, pp. 569-709, 1999.[116] M.  A. Jensen and J. W. Wallace, ?A review of antennas and propagation forMIMO wireless communications,? IEEE Trans. Antennas Propag., vol. 52,pp.2810-2824, 2004.[117] A. Frozena, D. J. Love, and R. W. Heath, ?Simplified spatial correlation models forclustered MIMO channels with different array configurations,? IEEE Trans. Veh.Technol.,vol. 56, pp. 1924-1934, 2007.[118] J. Molina-Garcia-Pardo, M. Lienard, A. Nasr, and P. Degauque, ?On the possibilityof interpreting field variation and polarization in arched tunnels using a model forpropagation in rectangular or circular tunnels,? IEEE Trans. Antennas Propag.,vol. 56, pp. 1206-1211, 2008.[119] J. Molina-Garcia-Pardo, M. Lienard, P. Degauque, C. Garcia-Pardo, and L. Juan-Llacer, ?MIMO channel capacity with polarization diversity in arched Tunnels,?IEEE Antenna Wireless Propag. Letters, vol. 8, pp. 1186-1189, 2009.193[120] Z. Sun and I. F. Akyildiz, ?Optimal MIMO antenna geometry analysis for wirelessnetworks in underground tunnels,? Proc. IEEE GLOBECOM, Nov. 30-Dec. 4,2009, pp.1-6.[121] V. R. Anreddy and M. A. Ingram, ?Capacity of measured Ricean and Rayleighindoor MIMO channels at 2.4 GHz with polarization and spatial diversity,? in Proc.IEEE WCNC, Apr. 2006, pp. 946-951.[122] G. Tsoulos, MIMO System Technology for Wireless Communications (ElectricalEngineering and Applied Signal Processing), CRC Press, Inc., 2006, p. 194.[123] S. F. Mahmoud, ?Modal propagation of high frequency electromagnetic waves instraight and curved tunnels within the earth,? J. Electromagn. Waves Appl., vol. 19,pp. 1611-1627, 2005.[124] I. Ben Mabrouk, L. Talbi, and m. Nedil, ?Performance evaluation of a MIMOsystem in underground mine gallery,? IEEE Antennas Propag. Lett., vol. 11,pp. 830-833, 2012.[125] T. Svantesson and J. Wallace, ?Statistical characterization of the indoor MIMOchannel based on LOS/NLOS measurements,? in Proc. IEEE Signals Syst.Computers, Provo, UT, USA, Nov. 2002, pp. 1354-1358.[126] A. Chehri, P. Fortier, H. Aniss, and P. M. Tardif, ?UWB spatial fading and smallscale characterization in underground mines,? IEEE Commun. Biennial Symp.,2006, pp. 213-218.[127] A. Chehri, P.  Fortier, and P. M. Tardif, ?Characterization of the ultra-widebandchannel in confined environments with diffracting rough surfaces,? WirelessPersonal Commun., vol. 62, pp. 859-877, 2012.[128] C. Garcia-Pardo, J. Molina-Garcia-Pardo, M. Lienard, D. P. Gaillot, andP. Degauque, ?Double directional channel measurements in an arched tunnel andinterpretation using ray tracing in a rectangular tunnel,? PIERM, vol. 22, pp. 91-107, 2012.[129] Remcom Incorporated, ?Wireless InSite User?s Manual?, version 2.6.3, PA, USA,2011.[130] K. D. Laakmann and W. H. Steier, ?Waveguides: Characteristic modes of hollowrectangular dielectric waveguides,? Appl. Opt., vol. 15, pp. 1334-1340, 1976.[131] S. M. Wentworth, ?Waveguides,? Applied Electromagnetics: Early TransmissionLines Approach, John Wiley, 2007, p. 344.[132] V. Erceg et al., ?IEEE P802.11 Wireless LANs,? TGn Channel Models, 2004.[133] L. Schumacher, K. I. Pedersen, and P. E. Mogensen, ?From antenna spacings totheoretical capacities-Guidelines for simulating MIMO systems,? in Proc. IEEEPIMRC, Sep. 2002, pp. 587-592.[134] B. Allen, R. Brito, M. Dohler, and H. Aghvami, ?Performance comparison of194spatial diversity array topologies in an OFDM based wireless LAN,? IEEE Trans.Consum. Electron. vol. 50, pp. 420-428, 2004.[135] H. Asplund, A. A. Glazunov, A. F. Molisch, K. I. Pedersen, and M. Steinbauer,?The COST 259 directional channel model - Part II: Macrocells,? IEEETrans. Wireless Commun., vol. 5, pp. 3434-3450, 2006.[136] K. I. Pedersen, P. E. Mogensen, and B. H. Fleury, ?Power azimuth spectrum inoutdoor environments,? Electron. Lett., vol. 33, pp. 1583-1584, 1997.[137] Q. H. Spencer, B. D. Jeffs, M. A. Jensen, and A. L. Swindlehurst, ?Modeling thestatistical time and angle of arrival characteristics of an indoor multipath channel,?IEEE J. Sel. Areas Commun., vol. 18, pp. 347-360, 2000.[138] C. X. Wang, X. Hong, H. Wu, and W. Xu, ?Spatial-temporal correlation propertiesof the 3GPP spatial channel model and the Kronecker MIMO channel model,?EURASIP J. Wireless Commun. Netw., 2007.[139] C. C. Chong, C. M. Tan, D. I. Laurenson, S. McLaughlin, M. A. Beach, and A. R.Nix, ?A new statistical wideband spatio-temporal channel model for 5-GHz bandWLAN systems,? IEEE J. Sel. Areas Commun., vol. 21, pp. 139-150, 2003.[140] R. M. Bueher, ?Generalized equations for spatial correlation for low to moderateangle spread,? Wireless Pers Commun., vol. 592, pp. 101-108, 2002.[141] R. Wonil and A. Paulraj, ?MIMO channel capacity for the distributed antenna,? inProc. IEEE VTC, Sep. 2002, pp. 706-709.[142] H. Zhang and H. Dai, ?On the capacity of distributed MIMO systems,? in Proc.IEEE CISS, Mar. 2004, pp. 1-5.[143] Z. Ni and D. Li, ?Effect of fading correlation on capacity of distributed MIMO,?Proc. IEEE PIMRC, Sep. 2004, pp. 1637-1641.[144] X. Wang, P. Zhu, and M. Chen, ?Antenna location design for generalizeddistributed antenna systems,? IEEE Commun. Lett., vol.13, pp. 315-317, 2009.[145] W. Choi and J. G. Andrews, ?Downlink performance and capacity of distributedantenna systems in a multicell environment,? IEEE Trans. Wireless Commun., vol.6, pp. 69-73, 2007.[146] A. M. Saleh, A. J. Rustako, and R. S. Roman, ?Distributed antennas for indoorradio communications,? IEEE Trans. Commun., vol. 35, pp. 1245-1251, 1987.[147] D. Wang, X. You, J. Wang, Y. Wang, and X. Hou, ?Spectral efficiency ofdistributed MIMO cellular systems in a composite fading channel,? in Proc. IEEEICC, May 2008, pp. 1259-1264.[148] C. Briso-Rodriguez, J. M. Cruz, and J. I. Alonso, ?Measurements and modeling ofdistributed antenna systems in railway tunnels,? IEEE Trans. Veh. Technol., vol. 56,pp. 2870-2878, 2007.[149] J. Kennedy and R. Eberhart, ?Particle swarm optimization,? in Proc. IEEE ICNN,195Nov.-Dec. 1995, pp. 1942-1948.[150] J. Robinson and Y. Rahmat-Samii, ?Particle swarm optimization inelectromagnetics,? IEEE Trans. Antennas Propag., vol. 52, pp. 397-407, 2004.[151] J. Nanbo and Y. Rahmat-Samii, ?Advances in particle swarm optimization forantenna designs: Real-number, binary, single-objective and multiobjectiveimplementations,? IEEE Trans. Antennas Propag., vol. 55, pp. 556-567, 2007.[152] M. Shihab, Y. Najjar, N. Dib, and M. Khodier, ?Design of non-uniform circularantenna arrays using particle swarm optimization,? J. Electrical Eng., vol. 59,pp. 216-220, 2008.[153] V. Erceg, P. Soma, D. S. Baum, and A. J. Paulraj, ?Capacity obtained frommultiple-input multiple-output channel measurements in fixed wirelessenvironments at 2.5 GHz,? in Proc. IEEE ICC, May 2002, pp. 396-400.[154] H. Sampath, S. Talwar, J. Tellado, V. Erceg, and A. J.  Paulraj, ?A 4th generationMIMO-OFDM broadband wireless system: Design, performance, and field trialresults,? IEEE Commun. Mag., vol. 40, pp.143-149, 2002.[155] R. W. Heath and A. J. Paulraj, ?Switching between diversity and multiplexing inMIMO systems,? IEEE Trans. Commun., vol. 53, pp. 962-968, 2005.[156] C. Oestges, ?Multi-link propagation modeling for Beyond Next Generationwireless,? in Proc. IEEE AP-URSI, Nov. 2011, pp. 1-8.[157] Agilent Technologies Inc., ?MIMO Performance and Condition Number in LTE,?Test Application Note, Oct. 2009.[158] Rohde & Schwarz, ?Assessing a MIMO Channel,? White Paper, Feb. 2011.


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



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


Related Items