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.

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Characterization of radiowave propagation in indoor industrial environments Stefanski, Adam 2010

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

Download

Media
24-ubc_2010_fall_stefanski_adam.pdf [ 2.03MB ]
Metadata
JSON: 24-1.0064889.json
JSON-LD: 24-1.0064889-ld.json
RDF/XML (Pretty): 24-1.0064889-rdf.xml
RDF/JSON: 24-1.0064889-rdf.json
Turtle: 24-1.0064889-turtle.txt
N-Triples: 24-1.0064889-rdf-ntriples.txt
Original Record: 24-1.0064889-source.json
Full Text
24-1.0064889-fulltext.txt
Citation
24-1.0064889.ris

Full Text

CHARACTERIZATION OF RADIOWAVE PROPAGATION IN INDOOR INDUSTRIAL ENVIORNMENTS by  ADAM STEFANSKI B.A.Sc., Simon Fraser University, 2004  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF APPLIED SCIENCE in  THE FACULTY OF GRADUATE STUDIES (Electrical and Computer Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA (Vancouver)  August 2010  © Adam Stefanski, 2010  Abstract In recent years, the successful introduction of short-range wireless technology in both consumer and commercial markets has attracted considerable interest from designers of industrial plants and factories. Wireless technology can be used to achieve both flexibility and cost reduction when installed and utilized for industrial process control networks and factory automation systems. Effective application of wireless devices in industrial environments requires careful assessment of the potential uses for such devices, methods for characterizing the wireless channel, and accurate models for the impairments introducted by the wireless channel. Here, we show that: (1) Although conventional wireless technologies such as ZigBee and WiFi are currently used in many industrial applications, ultrawideband (UWB) wireless technologies offer unique capabilities that may lead to their playing key roles in future industrial applications. (2) Our computerassisted technique for fitting the Saleh-Valenzuela model to measured UWB channel impulse responses (CIRs) offers a more rigorous and reproducable method for characterizing UWB channels than existing manual techniques can. (3) Although relatively little propagation data has been collected in indoor industrial environments to date, we combine these results to form a single rationalized UHF/UWB propagation model that is useful to designers and fills an important immediate need of designers while revealing gaps in our current understanding that need to be completed by future researchers.  ii  Table of Contents Abstract….….. .................................................................................................................................. ii Table of Contents ............................................................................................................................. iii List of Tables .................................................................................................................................... v List of Figures .................................................................................................................................. vi Acknowledgments .......................................................................................................................... vii Co-Authorship Statement .............................................................................................................. viii Chapter 1 Introduction..................................................................................................................... 1 1.1  References................................................................................................................. 5  Chapter 2 Applications of UWB Wireless Systems in Industrial Environments ............................ 7 2.2 2.2.1 2.2.2 2.3 2.3.1 2.3.2 2.3.3 2.3.4 2.4 2.4.1 2.4.2 2.4.3 2.5 2.5.1 2.5.2 2.5.3 2.5.4 2.6 2.7  Nature and Evolution of UWB .................................................................................. 8 Evolution of UWB ................................................................................................... 8 Classes of UWB Systems ........................................................................................ 9 The Industrial Wireless Environment ...................................................................... 10 Physical Environment ............................................................................................ 10 Networking Environment ...................................................................................... 10 Practical Issues and Challenges ............................................................................. 11 Wireless Standards and Advocacy......................................................................... 12 Industrial Applications of UWB .............................................................................. 12 Cable Replacement ................................................................................................ 12 Physical Sensors .................................................................................................... 14 Sensor Networks .................................................................................................... 16 Deployment Issues ................................................................................................... 19 Regulations and Standards ..................................................................................... 19 Channel Models ..................................................................................................... 22 Latency .................................................................................................................. 24 Security .................................................................................................................. 25 Conclusions ............................................................................................................. 25 References................................................................................................................ 27  Chapter 3 Computer-Assisted Identification of Clusters within UWB CIRs ................................ 34 3.1 3.2 3.3 3.3.1 3.3.2 3.3.3 3.3.4 3.4 3.4.1 3.4.2 3.5  Introduction.............................................................................................................. 34 Cluster Identification Approach .............................................................................. 36 Description of the Cluster Identification Algorithm................................................ 39 Local Smoothing in the Time Domain .................................................................. 41 Initial Search for Clusters ...................................................................................... 41 Iterative Search for Clusters .................................................................................. 42 Recursive Partitioning ........................................................................................... 43 Role of the Analyst .................................................................................................. 45 Generalized Cross-Validation Criterion ................................................................ 45 Additional Penalties ............................................................................................... 46 Validation ................................................................................................................ 47 iii  3.5.1 3.5.2 3.6 3.6.1 3.6.2 3.7 3.8  Simulated 802.15.4a CIRs ..................................................................................... 47 Measured UWB CIRs ............................................................................................ 51 Sources of Error ....................................................................................................... 55 Noise and Small-Scale Fading ............................................................................... 55 Overlap between Clusters ...................................................................................... 57 Conclusions ............................................................................................................. 58 References................................................................................................................ 60  Chapter 4 An Interim Propagation Model for Indoor Industrial Environments ........................... 62 4.1 4.2 4.3 4.3.1 4.3.2 4.3.3 4.3.4 4.4 4.4.1 4.4.2 4.4.3 4.5 4.6  Introduction.............................................................................................................. 62 Summary of Existing Measurements ....................................................................... 63 Rationalized Channel Parameters ............................................................................ 66 Classification of Propagation Topographies .......................................................... 66 Path-Loss and Shadowing...................................................................................... 67 Channel Impulse Response .................................................................................... 67 Small-Scale Fading ................................................................................................ 69 Results ..................................................................................................................... 70 Path-Loss and Shadowing...................................................................................... 70 Channel Impulse Response .................................................................................... 72 Small-Scale Fading ................................................................................................ 72 Conclusions ............................................................................................................. 73 References................................................................................................................ 74  Chapter 5 Conclusions and Recommendations ............................................................................. 76 5.1  References................................................................................................................ 78  Appendices ..................................................................................................................................... 79 Appendix A: Cluster Identification MATLAB GUI ............................................................. 79 Appendix B: UWB Channel Sounder .................................................................................... 84 Appendix C: UWB Channel Sounder MATLAB Code ........................................................ 89  iv  List of Tables Table 1: UWB Class Description .................................................................................................. 9 Table 2: List of 60 GHz Specifications and Standards ................................................................ 13 Table 3: Comparison of Wireless Technologies .......................................................................... 17 Table 4: Performance Comparison of Difference Harvesting Techniques .................................. 18 Table 5: Summary of UWB Channel Measurements in Indoor Industrial Environments ........... 24 Table 6: Number of Clusters Observed in Measured Responses ................................................ 53 Table 7: Extracted S-V Parameters for Measured UWB Responses ........................................... 55 Table 8: Summary of Indoor Industrial Propagation Measurement Campaigns ......................... 65 Table 9: Summary of Path-Loss Parameters ............................................................................... 70 Table 10: Summary of Wideband Channel Impulse Response Parameters ................................ 72 Table 11: Summary of UWB Channel Impulse Response Parameters ....................................... 72  v  List of Figures Figure 1: ISA100 Categories for Wireless Applications ............................................................... 11 Figure 2: International Regulation of WiMedia Alliance Band Groups........................................ 20 Figure 3: International Regulation of 802.15.4a Band Groups ..................................................... 20 Figure 4: International Regulation of High-Band UWB (57-64 GHz) .......................................... 21 Figure 5: Parameters of the (a) sparse multi-cluster (S-V) and (b) dense single-cluster channel impulse response models. ......................................................................................... 38 Figure 6: Flowchart of the rule-based cluster identification algorithm ......................................... 40 Figure 7: A locally smoothed power delay profile ........................................................................ 42 Figure 8: Evolution of the cluster identification process: (a) two clusters are broken into (b) three clusters, the second of which is broken on the next iteration to yield (c) four clusters. .................................................................................................................................. 44 Figure 9: Distribution of Clusters Among Simulated CIR ............................................................ 47 Figure 10: Estimation Accuracy using Simulated CIRs ................................................................ 48 Figure 11: Effectiveness of Algorithm for Simulated CIRs .......................................................... 48 Figure 12: RMS Error Threshold for Correct Cluster Identification ............................................. 49 Figure 13: Example of CIRs where (a) clusters #2, #4, and #12 could not be identified and (b) appears to have a positive slope ....................................................................................... 50 Figure 14: Representative Plots of CIRs and APDPs for CM3 ..................................................... 53 Figure 15: Representative Plots of CIRs and APDPs for CM7 ..................................................... 54 Figure 16: A PDP whose features have been buried by small-scale fading. ................................. 55 Figure 17: The components of the residual error........................................................................... 56 Figure 18: The ten identified clusters in this CM5 CIR are denoted by bold lines. The starts of the twelve actual clusters are marked by crosses ..................................................... 57 Figure 19: Plot of Line-of-Sight Path Loss vs. Distance ............................................................... 71 Figure 20: Plot of Obstructed Path Loss vs. Distance ................................................................... 71 Figure 21: Screenshot of the Initial Processing Step ..................................................................... 79 Figure 22: Screenshot of the Local Smoothing Step ..................................................................... 80 Figure 23: Screenshot of the Cluster Identification Step ............................................................... 81 Figure 24: Screenshot of the Final Adjustments Step ................................................................... 82 Figure 25: Block Diagram of the UWB Channel Sounder ............................................................ 84  vi  Acknowledgments This work was supported by grants from Omnex Controls, British Columbia Hydro and Power Authority, Western Economic Diversification Canada, and the Natural Sciences and Engineering Research Council of Canada. I thank my colleagues at the UBC Radio Sceience Laboratory, especially Shazhad Bashir, James Chuang and Shane Miller-Tait, for their many contributions to the work presented in this thesis. I also thank my supervisor, Prof. David G. Michelson, for encouraging me to take on this project and for his support and guidance during the course of this work.  vii  Co-Authorship Statement My thesis advisor, Prof. Michelson, and I jointly defined the organization and scope of the thesis research. Chapter 2 – I am the lead author of this chapter. Prof. Michelson played a key role in developing the outline, contributed important ideas, and reviewed the final result. Chapter 3 – I took over as lead author of this chapter when a previous MASc student, James Chuang, departed. I was responsible for extending the literature survey, collecting additional measurement data, extending the graphical user interface, and contributing major revisions to the text. Shane Miller-Tait helped resolve several software issues. Shazad Bashir assisted with validation of the technique. Prof. Michelson played a key role in developing the outline, contributed important ideas, and reviewed the final result. Chapter 4 – I am the lead author of this chapter. Prof. Michelson played a key role in developing the outline, contributed important ideas, and reviewed the final result.  viii  Chapter 1 Introduction Interest in using wireless communications in industrial environments has steadily increased over the past few decades, inspired to a large extent by the success of wireless technologies in home and office environments. In 2008, Venture Development Corporation reported that the market for wireless products used in industrial monitoring and control applications is expected to grow 19% annually for the next five years, reaching over US$1.5 billion in 2012 [1]. Other research groups, such as the ARC Advisory Group and IMS Research for example, forecast the annual growth over the next 5 years to be even larger at 25% and 28%, respectively. Typical existing industrial wireless applications include data acquisition systems, network products (access points, modems, repeaters, etc.), operator interface terminals (handhelds, tablets, vehicle mounts, etc.), remote controls, and sensors and transducers (flow, pressure, temperature, etc.). In general, strong growth rates are forecast in all of the aforementioned product categories, except for wireless remote controls due to the conservative nature of the application and the relative maturity of existing products. The many benefits of wireless, rather than wired, communication in industrial environments are presented in [1]–[3] and are a major reason for the strong market growth forecasts. Current and prospective users cite the lower installation costs as a top economic justification for switching to wireless communication since the installation and required maintenance of the large number of cables is greatly reduced. Industrial environments are typically harsh and chemicals, vibrations, and moving parts can potentially damage cabling. Additionally, the elimination of cables allows more flexibility for additions and changes to the factory floor. Also, wireless communication enables mobile systems and/or robots to communicate with fixed infrastructure, whereas with wired communication this would be impossible. Other benefits include ease of tracking assets, ease of implementation, and improved productivity and efficiency. One of the major challenges with communication in industrial environment is to ensure communication that is robust, secure, predictable, and reliable. Existing fieldbus systems, such as PROFIBUS [4], Modbus [5], CAN [6], are hard-wired and have been specifically 1  designed for industrial environments to provide reliable and predictable real-time communication. As expected, users mention that the most common problems experienced with wireless communication in industrial environments are RF interference and signal reception dropout and blockage. These issues need to be addressed since it jeopardizes system reliability and imposes the requirement for more access points. System implications and guidelines that enable users to realize better system planning, site assessment, and improved deployment of access points will greatly benefit users of wireless technology in industrial environments. Wireless technologies that do not require access to licensed spectrum are of particular interest in cost-conscious industrial environments. Bluetooth, which was published as the IEEE 802.15.1 standard, is a Wireless Personal Area Network (WPAN) standard that was developed as the original cable replacement technology [7]. Bluetooth operates in the license-exempt industrial, scientific, and medical (ISM) band at 2.4 GHz, utilizes adaptive frequency hopping spread spectrum to avoid interference, and has 79 channels of 1 MHz bandwidth each. The IEEE 802.15.4 Low-Rate WPAN standard became very attractive in industrial environments because of its low complexity, cost, and power consumption [8]. IEEE 802.15.4 systems can not only operate at the 2.4 GHz ISM band but also the 868 MHz ISM band in Europe and 915 MHz ISM band in North America. The 802.15.4 systems use Direct Sequence Spread Spectrum (DSSS) to avoid interference. This standardization effort led to the high-level specifications of Zigbee, which operates in the 2.4 GHz band has 16 channels with 5 MHz bandwidth each [9]. High-level specifications like WirelessHART [10] and ISA-100.11a [11] are being developed for 802.15.4 and are designed specifically for use in industrial environments. IEEE 802.11 technologies are also used in industrial environments [12]. 802.11a operates in the unlicensed 5 GHz band, whereas 802.11b and 802.11g operate in the 2.4 GHz band. 802.11b uses DSSS since it is an extension of the original 802.11 standard while 802.11a and 802.11g are based on orthogonal frequency-division multiplexing (OFDM) which allows for higher data rates. Lastly, per the recommendations of Task Group 802.15.4a, the IEEE amended the existing 802.15.4 standard with alternate physical layers at the UWB frequencies of 3.2 to 4.7 GHz (low-band) and 5.9 to 10.2 GHz (high-band). The performance and reliability of any wireless communication system is determined by the operational environment’s attenuation, fading, and time dispersion. For industrial 2  wireless communication, the propagation channel is much different than the home and office propagation channel. The industrial environment has very different obstructions and scatterers: factories have segmented equipment and complex structures rather than walls, signals experience much more multipath scattering, and on average the spaces and rooms are much larger. Factories usually have very few partitions other than external structural walls, whereas office and buildings commonly have many walls dividing rooms and offices. Furthermore, factories usually have lots of metallic instrumentation which causes multipath scattering and interference. For this reason, propagation models based on measured data from home and office environments are not directly transferable to industrial environments. Over the last twenty years, only a handful of researchers have presented measurement results, and subsequently proposed statistical models, for wireless propagation in indoor industrial environments. In [14]–[19], Rappaport and McGillem published a series of journal papers that were the first to detail wireless propagation in industrial environments. The papers were based on extensive UHF (1300 MHz) measurements collected in five factories in Indiana and included narrowband and some wideband results. With the emergence of UWB and its resilience to multipath interference, Karedal and Molisch published measurement results and a statistical channel model based on the measurements campaigns in two industrial halls [20]–[22]. The preliminary results from those campaigns were used by the 802.15.4a channel modeling sub-committee [23]–[24]. More recently, Tanghe presented results for spatial and temporal fading in the ISM [25] and UWB [26] bands. Clearly, the very limited number of measurements campaigns means the results are still preliminary and make it difficult to draw general site-independent conclusions. A comprehensive representation of the propagation channel will only emerge when a large body of measurement evidence is available. The work presented in this thesis represents three contributions to the characterization of radiowave propagation in indoor industrial environment: In Chapter 2, we present what we believe is the first comprehensive review of the potential applications for UWB wireless technology in indoor industrial environments. We conclude that UWB wireless devices and networks have the potential to play a significant role in industrial applications in the form of: (1) cable replacements, (2) physical sensors, 3  and (3) sensor networks. However, the lack of propagation models suitable for use in system design or evaluation increases risk for developers and is a significant impediment to progress. In Chapter 3, we demonstrate that efforts to characterize UWB channel impulse responses (CIRs) in new environments will benefit from a computer-assisted cluster identification algorithm for fitting measured CIRs to the Saleh-Valenzuela (S-V) model that is potentially more consistent and less time-consuming than current manual approaches. Cluster identification trials conducted using both simulated and measured UWB CIRs confirm the robustness and practicality of our approach. In Chapter 4, we present a rationalized propagation model that captures path loss and shadowing, fading and frequency dispersion, and time dispersion across the range 800 MHz to 10 GHz in selected deployment scenarios based upon the handful of propagation measurement campaigns that have been conducted to date in indoor industrial environments. The four deployment scenarios include both line-of-sight and non-line-ofsight paths in both heavy and light clutter. We also identify gaps in previous work that need to be filled. The results will be useful to both practitioners who require access to siteindependent models for analysis and design and to researchers who wish to contribute to the further development of such models. In Chapter 5, we summarize our contributions, assess the limitations of the work presented here and offer recommendations for future work.  4  1.1 References [1]  J. Taylor, The Worldwide Market for RF/Microwave Industrial Wireless Monitoring and Control Products for Discrete and Process Manufacturing. A White Paper On: Venture Development Corporation. Natick, Mass. Mar. 2008.  [2]  A.Willig, K. Matheus and A.Wolisz, “Wireless technologies in industrial networks,” Proc. IEEE, vol. 93, no. 6, pp. 1130–1150, Jun. 2005.  [3]  A. Willig, "Recent and Emerging Topics in Wireless Industrial Communications: A Selection," IEEE Trans. Ind. Informat., vol. 4, no.2, pp. 102-124, May 2008.  [4]  Specification on the General Purpose Field Communication System, 1996. [Online] Available: http://www.profibus.com/  [5]  Specification on the Modbus Application Protocol, 2006. [Online] Available: www.modbus.org/  [6]  ISO Standard 11898 – Road Vehicles – Controller Area Network (CAN), 1993. [Online] Available: http://www.iso.org/  [7]  Specification of the Bluetooth System, 2004. [Online]. Available: http://www.bluetooth.org  [8]  IEEE 802.15.4 standard, “Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR-WPANs)”, Oct. 2003.  [9]  Specification of the Zigbee System, 2007. [Online]. Available: http://www.zigbee.org/  [10] Specification of the HART Field Communication Protocol, 2007. [Online] Available: http://www.hartcomm.org/ [11] ISA 100.11a standard, “Wireless Systems for Industrial Automation: Process Control and Related Applications” 1st edition, 2009. [12] IEEE Standard Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) Specifications (P802.11), 2007 [13] IEEE 802.15.4a-2007 standard, “Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs); Amendment 1: Add Alternate PHYs," Mar. 2007. [14] T. S. Rappaport and C. D McGillem, “UHF multipath and propagation measurements in manufacturing environments,” in Proc. IEEE GLOBECOM, pp. 825-831, 1988. [15] T. S. Rappaport and C. D McGillem, “UHF fading in factories,” IEEE J. Sel. Areas Commun., vol. 7, no. 1, pp. 40-48, Jan. 1989. [16] T. S. Rappaport, "Indoor radio communications for factories of the future," IEEE Commun. Mag., vol. 27, no. 5, pp. 15-24, May 1989. [17] T. S. Rappaport, "Characterization of UHF multipath radio channels in factory buildings," IEEE Trans. Antennas Propag., vol. 37, no. 8, pp.1058-1069, Aug. 1989.  5  [18] T. S. Rappaport, S. Y. Seidel and K. Takamizawa, “Statistical channel impulse response models for factory an open plan building radio communication system design,” IEEE Trans. Commun., vol. 39, no. 5, pp. 794–807, May 1991. [19] P. Yegani and C. D. McGillem, "A statistical model for the factory radio channel," IEEE Trans. Commun., vol. 39, no. 10, pp. 1445-1454, Oct. 1991. [20] J. Karedal, S. Wyne, P. Almers, F. Tufvession and A. F. Molisch, "Statistical analysis of the UWB channel in an industrial environment," in Proc. IEEE VTC’04, pp. 81-85, Sep. 2004, [21] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "UWB channel measurements in an industrial environment," in Proc. IEEE GLOBECOM, pp. 35113516, Dec. 2004. [22] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "A MeasurementBased Statistical Model for Industrial Ultra-Wideband Channels," IEEE Trans. Wireless Commun., vol. 6, no. 8, pp. 3028-3037, Aug. 2007. [23] A. F. Molisch et al., “IEEE 802.15.4a Channel Model Final Report,” Tech. Rep., IEEE 802.1504-0062-02-004a, 2005. [24] A. F. Molisch et al., "A Comprehensive Standardized Model for Ultrawideband Propagation Channels," IEEE Trans. Antennas Propag., vol. 54, no. 11, pp. 31513166, Nov. 2006. [25] E. Tanghe, W. Joseph, L. Verloock, L. Martens, H. Capoen, K. Van Herwegen and W. Vantomme, "The industrial indoor channel: large-scale and temporal fading at 900, 2400, and 5200 MHz," IEEE Trans. Wireless Commun., vol. 7, no. 7, pp. 27402751, Jul. 2008. [26] E. Tanghe, W. Joseph, J. De Bruyne, L. Verloock and L. Martens, "The Industrial Indoor Channel: Statistical Analysis of the Power Delay Profile,” in Int. J. of Elect. Commun., vol. 64, no. 9, pp. 806-812, Sep. 2010.  6  Chapter 2 Applications of UWB Wireless Systems in Industrial Environments1 2.1  Introduction  Short-range wireless technology is a proven method for: (1) simplifying connection to mobile and rotating assets, (2) reducing cabling and re-configuration costs, and (3) substantially reducing the risk of cable or connector damage. Although conventional wireless technologies such as ZigBee and WiFi are currently used in many industrial applications, ultrawideband (UWB) wireless technologies offer unique capabilities that may lead to their playing key roles in industrial applications in the future. In particular, UWB wireless technology offers: (1) greater resistance to multipath fading and interference, (2) potentially lower power consumption, (3) support for higher data rates, and (4) provision for accurate measurement of distance, velocity and position, and intrusion. Although it seems likely that UWB wireless devices and networks have the potential to play a significant role in industrial applications in the form of: (1) cable replacements, (2) physical sensors, and (3) sensor networks, researchers have been hampered by the lack of a thorough review of previous work in this area that would help them identify gaps to be filled and put their own work in context. Accordingly, here we review previous efforts to assess or deploy UWB wireless systems in indoor environments. The remainder of the chapter is organized as follows: in Section 2.2, we present the nature and evolution of UWB communications and discuss the two classes of UWB systems. Section 2.3 describes in detail the industrial wireless environment. We explain the physical difference between office buildings and factories, the industrial networking and connectivity environment, and present the organizations promoting the user of wireless technologies in industrial environment. Section 2.4 describes in detail the industrial applications for UWB. Finally, in Section 2.5 we discuss deployment strategies, channel models, security, and latency consideration for UWB wireless systems operating in industrial environments. A summary and concluding remarks conclude the chapter. A version of this chapter has been presented at the Ultra-Wideband Technology: The State-of-the-Art and Applications workshop at IEEE MTT 2010 International Microwave Symposium A version of this chapter will be submitted for publication: A. Stefanski and D.G. Michelson, “Applications of UWB Wireless Systems in Industrial Environments.”  7  2.2  Nature and Evolution of UWB  UWB communication systems have numerous potential advantages over narrowband and wideband wireless systems that make UWB very attractive for operation in industrial environments, such as (1) the very fine time resolution (< 2ns) which allows for accurate distance, position, and velocity measurements, (2) greater immunity to noise, interference and/or jamming due to the wide bandwidth, (3) greater resistance to multipath fading, (4) potentially higher data rates, (5) lower power consumption, and (6) increased security. The low power results in communication that is very secure since the signal will be very close to the noise floor. Furthermore, the wide bandwidth allows for a large processing gain and enables the signal to carry reliably through the multipath-rich industrial environment, whereas narrowband signals may experience serve fading and outage. Also due to its very wide bandwidth, UWB is not easily susceptible to jamming from higher powered narrowband signals. Finally, UWB can transmit very high data rates (i.e., for data acquisition systems) and also provide precision asset tracking and positioning as a result of the fine spatial resolution which is inversely proportional to bandwidth. 2.2.1  Evolution of UWB  The start of contemporary UWB theory2 is based on the radar impulse work done by Dr. Gerald Ross at Sperry Research Center and Dr. Harmuth at Catholic University in the 1960s and 1970s. Since the late 1960s, UWB signals have been used in high-resolution military radars. Non-military UWB gained prominence when [3]–[7] published work on time-hopping impulse radio and UWB propagation measurements in the late 1990’s. In 2001, Molisch and Win published the first statistical model for the UWB indoor channel based on a stochastic tap-delay line [8]. The work showed that the power delay profile of a typical office building can be modeled by an exponential time constant and included path loss analysis. Since then, there has been extensive research done into UWB communication systems. A communication theorist may define UWB as a limiting case strategy for utilizing a signal’s time-bandwidth-power product in order to achieve fine time resolution, resistance 2  At the beginning of the 20th century, Hertz and Marconi developed a spark-gap transmitter which created an impulse excitation of an antenna which produced a frequency spectrum similar to UWB. However, the interest in the field was relatively dormant and contemporary UWB is much different than the initial experiment.  8  to fading and interference, low probability of intercept and, potentially, high data rates. In 2002, the Federal Communications Commission (FCC) in the USA was the first international jurisdiction to formally regulate UWB and the FCC defined UWB signals in terms of transmission bandwidth from an antenna where the emitted signal has an absolute bandwidth greater than 500 MHz or with a fractional bandwidth greater than 20% of its centre frequency (whereas wideband signals only have 1% to 20% fractional bandwidth). Other jurisdictions followed with their own definitions of UWB by mainly by varying the minimum required bandwidth between 50 MHz (EU), 450 MHz (Japan), and 500 MHz (Korea, Canada). 2.2.2  Classes of UWB Systems  From a regulatory perspective, civilian UWB systems typically fall into one of two classes as defined by the FCC: (1) above 40 GHz which is covered by FCC Report & Orders 94-499 and 98-150, and (2) below 30 GHz which is covered by FCC Report & Order 02-48. UWB above 40 GHz, with emphasis on the license-exempt 57-64 GHz band, was first approved by the FCC in 1995 and only recently approved in Europe. For technical reasons, commercial exploitation has only recently begun. UWB below 30 GHz, with emphasis on the license-exempt bands below 960 MHz and between 3.1 GHz and 10.6 GHz, was the last to be approved by the FCC and a combination of technical and regulatory hurdles have impeded both progress and adoption. TABLE 1: UWB CLASS DESCRIPTION Name  Freq. Range  Authorized Applications  Power Restrictions  Low-band UWB  < 960 MHz  Imaging  Low power (-41 dBm)  Mid-band UWB  3.1 – 10.6 GHz  Imaging, communication, and measurement  Low power (-41 dBm)  High-band UWB  57 – 64 GHz  Communication and measurement  High power (+10 dBm)  9  2.3  The Industrial Wireless Environment  Coupled with the global success of wireless technology in the consumer market, the use wireless technology in industrial environments is very attractive because of the potential for cable cost savings, greater plant/factory flexibility, possible location and tracking of assets, and increased process optimization [9-10]. However, it is first necessary to understand the nature of the industrial wireless environment since the physical and network properties are significantly different than the residential and commercial office buildings where consumer products operate. 2.3.1  Physical Environment  As described by Rappaport and McGillem in [11]-[16], there are gross differences between office buildings and factories. The typical industrial plant comprises of a relatively small control room surrounded by a relatively large physical plant and/or factory floor. The environment can be characterized by large open spaces with orthogonal aisles and considerable shadowing from machinery and manufacturing equipment, confined spaces with signal blockage and shadowing, and rotating and/or mobile machinery. The ceiling typically has metal trusses and is much higher than in office buildings. The unique construction and floor arrangement of industrial factories is significantly different than office buildings, hence, the radiowave propagation is also considerably different. Industrial environments also typically suffer from harsh temperature, vibrations, and exposure to chemicals. The numerous metal machinery and moving components create much more electromagnetic interference and noise compared to the typical consumer environment, however, most of noise signatures decay rapidly above 1 GHz [13]. Wireless devices operating in industrial environments also have to contend with higher levels of multipath interference which is created by multiple reflects from metal machinery and structures. Therefore, deployment strategies of wireless communication systems must account of the relative severity of the industrial environment. 2.3.2  Networking Environment  As with the physical environment, the industrial networking environment is much different than the commercial office environment. A thorough overview is provided in [17]. The control room communicates with local controllers, programmable logic controllers 10  (PLCs), and sensors and actuators throughout the plant. The connectivity is typically provided by dedicated wires, fieldbus communication systems, or industrial Ethernet. The control room and the plant mostly exchange state information, hence, most of the control is supervisory. This means the throughput is generally low but security, reliability, and realtime requirements are paramount. The International Society of Automation (ISA) has proposed six roles that wireless technology can fill in industrial networks, as summarized in  Category  Class  Safety  0 1  Control  2  Application  Description  Emergency action  (always critical)  Closed loop regulatory control Closed loop supervisory control  3  Open loop control  4  Alerting  5  Logging and downloading/uploading  Monitoring  (often critical) (usually non-critical) (human in the loop) Short-term operational consequence (e.g., event-based maintenance) No immediate operational consequence (e.g., history collection, sequence-of-events, preventive maintenance)  Importance of message timeliness increases  Figure 1 [18].  Figure 1: ISA100 Categories for Wireless Applications 2.3.3  Practical Issues and Challenges  As with most new technologies, there are both practical issues and design challenges that must be considered and understood. As described in [19], the major technical challenges of wireless industrial networks are: 1) resource constraints, 2) dynamic topologies, 3) harsh environmental conditions, 4) quality-of-service (QoS) requirements, 5) data redundancy, 6) security, and 7) network integration. Also, an increasing number of network components are powered over Ethernet (PoE), or using similar schemes that run data and power in parallel over the same cable; eliminating the cable will eliminate the power. With these challenges in mind, industrial and academic researchers are developing more robust wireless system architectures since the physical, media access control, and routing layers all influence the contention for available network resources. Nonetheless, wireless technology holds great potential for industrial and factory automation. The many benefits of wireless, rather than wired, communication in industrial environments are presented in [9], [10], and [20]. Foremost, current and planned users cite 11  the lower installation costs as the top justification for switching to wireless communication since the installation and required maintenance of the large number of cables, and risk of potentially cable damage, is greatly reduced. The elimination of cables also allows more flexibility for additions and reconfigurations of the factory floor. Wireless communication also enables mobile systems and/or robots to communicate with stationary infrastructure, whereas with wired communication this would be impossible. Other benefits include ease of tracking assets, ease of implementation, simplified health monitoring of equipment, and improved productivity and efficiency. 2.3.4  Wireless Standards and Advocacy  Worldwide, there are several organizations promoting the use wireless technologies in industrials environments. The IEEE Industrial Electronics Society (IES) is devoted to the application of electronics and electrical sciences for the enhancement of industrial and manufacturing processes [21]. The IEEE IES publishes periodicals documenting the stateof-the-art in industrial electronics, organizes conferences to foster communication between practitioners and the research community, and actively participates in committees for formal wireless standards such as 802.11, 802.15, and 802.16. Similarly, the International Society of Automation (ISA) is a non-profit organization consisting of industry professionals and is active in all aspects of industrial and factory automation [22]. The ISA develops standards, hosts conferences and exhibitions, publishes books and technical articles, and also provides training and education. Lastly, the Wireless Industrial Networking Alliance (WINA) primarily focuses on education and advocacy of wireless industrial networking with the mission to accelerate widespread deployment of wireless technologies in industrial environments [23].  2.4  Industrial Applications of UWB  2.4.1  Cable Replacement  2.4.1.1 Mid-band UWB (3.1-10.6 GHz)  After the FCC regulated UWB in 2002, academic and industry research had considerable interest in the technology’s potential as a short-range and high data-rate communication system since UWB offered very large bandwidths, resiliency to multipath fading, and low 12  interference [24]. The IEEE established a task group to develop alternate UWB physical layers (PHY) for the 802.15.3 high-rate WPAN standard, however, the Task Group 3a was disbanded in 2006 because, after down-selecting to two proposals, it was in a deadlock between Multiband Orthogonal Frequency Division Multiplexing (MB-OFDM) supported by the WiMedia Alliance and Direct Sequence Spread Spectrum (DSSS) supported by the UWB Forum [25]. Consequently, Motorola and Freescale, two of the three founders, pulled out of UWB Forum and the forum become defunct shortly thereafter. The WiMedia Alliance continued to promote the adaptation and standardization of UWB worldwide and to establish a compliance, certification, and interoperability program. As a result of their efforts, ECMA International released two ISO-based standards, ECMA-368 [26] and ECMA-369 [27], founded on WiMedia’s MB-OFDM platform. Currently, the WiMedia Alliance has more than 350 members and is working on a 1024 Mbps version of their Common Radio Platform specification for streaming media applications. 2.4.1.2 High-Band UWB (57-64 GHz)  The FCC also allocated the 57-64 GHz license-exempt band for wireless high data-rate communications. The benefits of moving to a millimetre wave band allow for high datarates using low costs, high gain antennas, and high component integration. However, there are challenges with path-loss due to the short wavelength, oxygen absorption at long distances, and obstruction since directional antennas can be easily obstructed due to the narrow beamwidth. Another challenge for industry adaptation is the number of competing specifications, which are listed in Table 2. TABLE 2: LIST OF 60 GHZ SPECIFICATIONS AND STANDARDS Name WirelessHD  Version 1.0  Release Jan. 2008  Data Rate up to 4 Gbps  ECMA-387  1.0  Dec. 2008  up to 6.35 Gbps  IEEE 802.15.3c  802.15.3c-2009  Oct. 2009  up to 5.28 Gbps  Wireless Gigabit Alliance  1.0  Dec. 2009  up to 7 Gbps  IEEE 802.11ad  -  ~Dec. 2012  to be determined  13  2.4.2  Physical Sensors  2.4.2.1 Localization and Tracking of Assets  Precise tracking of assets has becoming increasingly important in industrial applications as decreasing profit margins and increased competition are propelling operations to achieve optimal efficiencies. Real-time location systems (RTLS) help ensure that assets are in the right place at the right time [28]. There are numerous possible applications afforded by tagging assets such as process automation, personnel identification/tracking, equipment tracking, managing inventory, etc. RTLS provides valuable information that enables companies have increased visibility and control of their manufacturing process, eliminate inefficiencies, and improve security. UWB technology inherently enables the accurate localization of assets.  Unlike  traditional wireless ranging systems which use receive signal strength indication (RSSI) to determine location, UWB can exploit time-based ranging since the short temporal pulses used by UWB result in a very fine spatial resolution. Time-based systems measure the propagation delay and apply time-of-arrival (TOA) techniques, or variants such as timedifference-of-arrival (TDOA), to determine the location. Practical UWB estimators in multipath environments can easily resolve the root-mean-square error of the arrival pulse to less than 1ns which is equivalent to approximately 30cm, depending on the bandwidth and signal-to-noise ratio. [29] provides an thorough overview of the theory of ranging with UWB in multipath environments and also highlights some of the challenges and sources of estimation error. 2.4.2.2 Commercial UWB Real-Time Locating Systems  As summarized in [30], there are commercially available UWB RTLS’s from Sandlinks (Total Asset Visibility), Time Domain (Plus® RTLS), Ubisense, and Zebra Enterprise Solutions (Sapphire). Generally, these systems have comparable performance specifications which include sub-30cm location accuracy, operations distances of up to 50m indoors (LOS), refresh rates up to 30 Hz, and tag battery life of 4-7 years. The systems typical operate between 6 GHz to 8 GHz (region dependant) and have approximately 1 GHz of bandwidth, other than Sandlinks which operates at 4 GHz with only 500MHz bandwidth. All of the systems are utilizing a proprietary protocols, therefore, they are not interoperable 14  and only function with the respective manufacturer’s software application. 2.4.2.3 Ranging in the IEEE 802.15.4a Standard  The lack of device interoperability has hindered market proliferation, however, the ratification of the 802.15.4a standard in 2007 should rectify this [31]. The IEEE 802.15.4a standard is the first international standard that specifies a physical layer with a precision ranging capability and should enable a wide range of standard-based RTLS and ultralow power applications [32]. Recently, DecaWave announced the first 802.15.4a compliant UWB device. As per the 802.15.4a standard, the device is able to operate as a full-function device (FFD) where accuracy is critical or as a reduced-function device (RFD) were power consumption is paramount [33]. The transceiver has 10 cm location accuracy at a line-ofsight range of 70 m range, even when moving at speeds up 5 m/s. It is also capable of transmitting data at 6.8 Mbps or operating from a single battery for up to 10 years. Although the device can only operate between 3.5 GHz to 7 GHz (i.e., bandgroups 1-7), it is compatible with the regulations in major international markets (USA, EU, and Japan). 2.4.2.4 802.15.4f and RFID  Decawave, along with the aforementioned commercial UWB RTLS companies, are also all currently working towards a new IEEE 802.15 standard for active radio frequency identification (RFID) systems. The IEEE 802.15.4f task group is chartered with defining a new PHY layer, along with the necessary enhancements to the 802.15.4 standard MAC layer, to support active RFID system for bi-directional and location determination applications [34]. The new standard will specify three new PHYs (UHF, 2.4 GHz, and UWB). For UWB PHYs, a new frequency plan will use existing specified 802.15.4a bands to promote interoperability, however, it will not use the 4a mandatory bands to avoid interference. Furthermore, the minimum bandwidth for the new 4f bandgroups will be decreased to 400 MHz while the maximum bandwidth will include multiple 4a bands to ensure low-cost filtering methods can be used [35]. Obviously, having an international RFID standard that allows the passage of active RFID tags across international borders will enable the largest market penetration.  15  2.4.2.5 Surveillance Systems  Another potential UWB application outlined by the FCC is surveillance systems [24]. It is possible to use UWB systems to establish an RF perimeter fence (i.e., “a security fence”) that can detect intrusion since UWB technology affords very good clutter rejection. A lowcost UWB device can be used to activate higher performance imaging equipment upon detection. [36] summarizes the recent advances in UWB radar and ranging systems designed for perimeter intrusion detection, obstacle and collision avoidance, and industrial safety applications and [37] outlines the capability of cooperative UWB anti-intruder radar systems. The technology is been commercially available and Time Domain has recently released a new P400 UWB Radar Sensor module which has a small 3”x3” form factor, is self-configuring, and had a 90 m detection range with 2 cm resolution [38]. 2.4.2.6 Novel Applications  There are also many novel UWB sensors that have potential applications in industrial environments. It is now possible to use low-cost UWB sensors for detection of foreign objects in logs, food, etc. along production lines, which is much less expensive than x-ray and gamma radiation and does not require ionization [39]. Another application is using low-cost UWB patch-antenna sensors, rather than high-frequency current transformers (HFCT), to detect partial discharge of stator windings in high voltage machines for quality assurance and early maintenance [40]. It is also possible to use UWB for fluid-level sensing licensing using Lawrence Livermore National Laboratory’s (LLNL) patented lowpower UWB micropower impulse radar [41]. More recently, Xsens and Time Domain announced a state-of-the-art 3D motion capture sensor that combines inertial sensor technology with UWB precision location technology to enable scalable multi-person motion capture systems that do not have line-of-sight restrictions. 2.4.3  Sensor Networks  2.4.3.1 Overview The competitive industrial marketplace requires improved productivity and efficiency to meet corporate objectives. Industrial automation sensor networks that intelligently collect and process system information are required, however, wired systems have prohibitive 16  cable installations and high maintenance costs. Industrial wireless sensor networks offer several key advantages over traditional wired monitoring, such as low installation and maintenance costs, self-organization, rapid deployment, system flexibility and ease of reconfiguration, and intelligent processing capabilities [19]. Industrial automation systems typically consist of three components [42]. Low complexity wireless sensors monitor critical physical parameters such as vibration, temperature, pressure, power quality, voltage, and can also send the sensor’s position. The wireless sensor nodes convey the information to access points (AP) where the data is analyzed or forwarded via a backbone network. Although the traffic generated by the sensor node typically has a relatively low data-rate and is unidirectional, the AP must be able to cope with aggregated traffic and multiple access. Finally, plant personnel are notified of potential problems through an advanced warning system and are able to address the issues before efficiency drops and/or equipment needs to be replaced. 2.4.3.2 Benefits of UWB Sensor Networks  When considering the deployment of sensor networks in industrial environments, it is necessary to examine the benefits of using UWB over existing wireless technologies such as Zigbee and WiFi. As described in [43] and highlighted in Table 3, UWB offers significant advantages over other wireless systems since it accommodates high data-rates at very low powers, is resilient to multipath fading, and provides excellent location capabilities. Furthermore, the challenges with hardware development, MAC suitability, multipath interference, and modeling the propagation environments are starting to be more thoroughly understood and industrial UWB sensors networks are already realizable. TABLE 3: COMPARISON OF WIRELESS TECHNOLOGIES 2.4 GHz ZigBee  2.4 GHz WiFi  UWB  Data Rate  low (250 kbps)  high (10-100 Mbps)  Medium (1-27 Mbps)  Transmission Distance  short (<30m)  long (up to 100m)  short (<30m)  Power Consumption  low (20-40mW)  high (0.5-1 W)  low (30mW)  Multipath Performance  poor  poor  good  Interference Resilience  low  medium  Interference to Others Complexity and Cost  high low  high high  high (with complex rcvr.) low (with simple rcvr.) low low/med/high  17  2.4.3.3 UWB applications  There are many applications for UWB sensor networks in industrial environments as described in [17], [19], [44]–[47].  Industrial environments not only require data  communication but also require localization and tracking of assets. This information allows for increased process control and efficiency. It also can reduce asset maintenance costs and increases workplace safety. Using the factory floor as an example: factories typically use separate systems for logging equipment location, updating production floor status, and tracking employees and visitors, however, UWB offers the unique advantage that it is able to accomplish all of the tasks using a single system. 2.4.3.4 Energy Harvesting  With the low power consumption of UWB sensors, one research area that is gaining interest is energy harvesting and it well reviewed in [48]. The use of batteries to power sensor networks can be a burden since, sooner or later, the battery will have to be replaced. However, it is possible to prolong the life of a sensor using techniques that harvest energy from the surrounding environment. Table 4 from [49] shows the required transducer size to achieve 10mW of power which, for example, is 10x the power required by new ultra-lowpower UWB chipsets developed for sensor network applications [50]. TABLE 4: PERFORMANCE COMPARISON OF DIFFERENCE HARVESTING TECHNIQUES Secondary Commercially Available Storage yes  Required Dimension -  Energy Source  Performance  Primary Battery  2880 J/cm3  Secondary Battery  1080 J/cm3  -  yes  -  Light (indoor)  10-100 µW/cm2  yes  yes  59 – 590 cm2  Airflow  0.4-1 mW/cm3  yes  no  6-15 cm3  Vibrations  200-380 µW/cm3  yes  yes  16-30 cm3  Thermoelectric Electromagnetic Radiation  40-60 µW/cm2  yes  yes  98-148 cm2  0.2-1 mW/cm2  yes  yes  6-30 cm2  18  2.4.3.5 IEEE 802.15.4a Sensor Networks  The IEEE 802.15.4a standard has been developed, in part, specifically for UWB sensor networks. As described in [51], the IEEE 802.15.4a standard defined a physical layer that provides the improved sensor network communication capabilities of 802.15.4 (currently used by Zigbee, Wireless HART, ISA 100.11a, etc.) and also provides sub-meter geolocation. As in 802.15.4, the amended 4a standard keeps the same distinctions between FFDs and RFDs and maintains support for star, cluster tree, and mesh network topologies [43].  2.5  Deployment Issues  In the following sections, we identify and discuss some of the key issues regarding the deployment of UWB wireless devices in industrial environments. 2.5.1  Regulations and Standards  International regulation and frequency allocation is fundamental to the commercialization and deployment of UWB wireless system in industrial environments. Standards committees, such as the IEEE and ISA, help develop application specific protocols to aid proliferation and interoperability. However, it is worth noting, that standardization does not always guarantee success in the commercial marketplace. The following sections discuss the situation, at the time of writing, for UWB wireless technology. 2.5.1.1 International Regulation of Mid-band UWB (3.1-10.6 GHz)  In 2002, the FCC was the first to regulate mid-band UWB and defined it as any signal that having fractional bandwidth greater than 20% or absolute bandwidth greater than 500 MHz [24]. Other countries followed with their own regulations, however, all were far more cautious. The EU requires UWB signals to have more than 50 MHz bandwidth but limited the main operational band to 6.0 GHz to 8.5 GHz with a restricted band, requiring detection and avoidance (DAA) and low duty cycles (LDC), between 3.1 GHz to 4.8 GHz and 8.5 GHz and 9.0 GHz [52]. In Japan, UWB is defined at having more than 450 MHz bandwidth and can be used at frequencies from 3.4 GHz to 4.8 GHz (with DAA) or 7.25 GHz to 10.25 GHz. In Korea, the frequency is restricted to 3.1 GHz to 4.8 GHz (with DAA) or 7.2 GHz to 10.2 GHz. In China, regulation only allows the use of UWB between 4.2 GHz and 4.8 19  GHz (with DAA) or between 6.0 GHz to 9.0 GHz approximately. Industry Canada recently announced that its UWB frequency spectrum will be limited to the 4.75 GHz to 10.6 GHz to protect Canadian radio services and C-band fixed satellite service communication [53]. Figure 2 and Figure 3, respectively, illustrate how the above international regulations relate to the band groups defined by the WiMedia Alliance and 802.15.4a Standard.  Figure 2: International Regulation of WiMedia Alliance Band Groups  Figure 3: International Regulation of 802.15.4a Band Groups 20  2.5.1.2 International Regulation of High-band UWB (57-64 GHz)  Unlike the Mid-Band UWB, the regulation of the license-exempt 60 GHz spectrum is far more consistent worldwide, as shown in Figure 4. Typically, the transmit power is limited to +10 dBm. However, the regulation allow designers to use directional antennas to achieve effective isotropic radiated power (EIRP) that are usually 30 to 40 dB higher [77].  Figure 4: International Regulation of High-Band UWB (57-64 GHz) 2.5.1.3 Standards-based vs. Proprietary Products  For developers of information and communications technology, jointly developed standards play an important role in building consumer demand for new products. Cooperative development helps to ensure that the needs of a broad community of potential users are met and the demand will be sufficient to allow economies of scale to be achieved. Standardization helps to ensure that: 1) products from multiple vendors will interoperate and 2) consumers can avoid vendor lock-in. The success of cooperatively developed standards such as IEEE 802.15.1 (Bluetooth), IEEE 802.11b (WiFi) and IEEE 802.15.4 (ZigBee) in creating whole market sectors in a short time has led to many similar wireless standardization efforts being undertaken. For UWB, however, cooperatively developed standards have not led to the success that their proponents anticipated. For high data-rate UWB, the WiMedia Alliance has developed a certification program for its UWB Common Radio Platform and, at its peak, there were 14 startups vying for market share. Currently, however, the number is much smaller with 21  devices only available from companies such as Alereon, Realtek Semiconductor, and Wisair. Decawave recently released the first and, as of yet, only single chip, 802.15.4a compliant device. However, it only supports the band groups between 3.5 GHz and 7 GHz. At the time of writing, the only standards-based devices available for 60 GHz UWB are WirelessHD-compliant transceivers from SiBeam. Compared to standards-based products, proprietary UWB products have achieved much greater commercial success. This can likely be attributed to the skill of the independent developer in correctly identifying a specific need that can be met using UWB techniques and in developing a solution that is well-matched to that need. In such cases, standardization efforts often begin after a proprietary product achieves commercial success. Sandlinks, Time Domain, Ubisense, and Zebra Enterprise Solutions are among the firms that offer proprietary UWB devices that serve specialized applications. 2.5.2  Channel Models  All wireless signals are affected by physical mechanisms (i.e., reflections, scattering, etc.), hence, measurement-based channel models are used to succinctly capture our understanding of the wireless propagation environment in a manner that is useful in the design, test, and simulation of communication systems [54]. High-quality statistical channel models not only help system designers predict system reliability and performance but they also allow designers to efficiently plan deployment. Channel models reduce the duration of the expensive field testing campaigns, reduce redesign cycle, and provide insight into why a system may not work as anticipated. 2.5.2.1 Mid-Band UWB (3.1-10.6 GHz) Industrial Channel Models  UWB propagation and channel models are fundamentally different than wideband models.. As UWB systems operate with a very large fractional bandwidth, one can no longer assume that pathloss is constant with frequency. Plus, with only a few components per delay bin the Central Limit Theorem, and its implication of complex Gaussian fading, is no longer valid [55]. Therefore, the fine resolution and complex structure of the channel impulse response must be considered and it is necessary to accurately characterize path loss, time dispersion, small scale fading, and other channel parameters and statistics. As previously mentioned, the first propagation measurements in industrial environments 22  were reported by Rappaport in [11] and were subsequently used for the results presented in [12]–[16]. The propagation measurements were taken at five fully operational factories that varied in size and layout and the separation distance was varied between 10m and 80m. A decade later, measurement results for three separate campaigns in the ISM band were published in [56]–[59]. The first results of UWB in industrial environment where presented in [60]–[63] and were based on very limited measurement campaigns. This work proposed an improved statistical channel model for industrial environments based on the modified Saleh-Valenzuela statistical model. The authors found that the energy arrives in clusters and that the large quantity of metallic scatters caused large amounts of multipath meaning that the number of multipath components required to capture most of the energy is quite large. The preliminary results from those campaigns were used by the 802.15.4a channel modeling committee [64]–[65]. Later, [66] developed a simple, yet flexible, locallycoherent UWB model for industrial environments that generalized the geometry-based stochastic channel model, a concept that is widely used in wideband channel models. More recently, [67]–[72] presented results for large and small scale statistics from UWB measurement campaign within different industrial environments. A summary of the five different UWB measurement campaigns in industrial environments is provided in Table 5. The limited set of available UWB measurement data in industrial environments means the results are still anecdotal since most measurement campaigns only occur at the single industrial site. A more comprehensive representation of the propagation channel will only emerge when a larger body of measurement data is available.  In particular, more  measurements are required to determine propagation in different classes of industrial environments, the frequency dependence through sub-band analysis, and distance dependence at longer ranges. Further research and improved models will enable a more complete understanding of UWB propagation in industrial environments and will allow designers to reliably predict system performance and plan more efficient deployments. 2.5.2.2 High-Band UWB (57-64 GHz) Industrial Channel Models  UWB propagation at 60 GHz requires different channel models than at 3.1 to 10.6 GHz. At 60 GHz, wireless signal experience much higher pathloss due to the short wavelength which causes severe material attenuation and shadowing of the diffraction region, hence, the usable range is limited to 10m or so in indoor environments. Furthermore, the time- of23  TABLE 5: SUMMARY OF UWB CHANNEL MEASUREMENTS IN INDOOR INDUSTRIAL ENVIRONMENTS UWB Channel Measureme nts  Ref.  Freq. (GHz)  Distance  Environment  # of Locations  Parameters Extracted  Karedal (2007)  [60][63]  3.1 - 10.6  2 - 16 m (LOS & NLOS)  Incinerator hall and factory hall  < 25  PL, PDP, delay spread, MPC, S-V model parameters  Kunisch & Pamp (2006)  [66]  3 - 11  < 20m (LOS) & <10m (NLOS)  Hall w/ some machinary  ?  PL, PDP, delay spread  Irahhauten (2006)  [67]  0.1 to 12  1m to 25m  Process industry laboratory  73  PL, delay spread  Yasmeen (2008)  [68][70]  3.1 - 10.6  4m to 10m  Laboratory facility  ?  PDP, delay spread  Tanghe (2009)  [71][72]  0.8 - 4.0  10m to 35m  Wood processing facility  5 LOS & 5 NLOS  PDP, S-V model parameters  arrival and angle-of-arrival are critical since the antennas are highly direction. To date, the only measurement-based 60 GHz channel models available are for indoor environments (presented in [73]–[81]) and there have not be any measurement campaigns in industrial environment. 2.5.3  Latency  Another phenomena associated with wireless communication system is latency, or time loss. Wireless transceivers are typically half-duplex since a transmitting node would saturate any receive signal if using the same channel [9]. Hence, most transceivers share circuitry to reduce size and cost, thereby, making concurrent transmission impossible. Also, wireless transceivers typically have larger physical layer overheads than wired system. A larger preamble is required at the start of a packet for carrier-/bit- synchronization to compensate for noisy channels. Finally, the propagation channel can also result in slow and fast fading which can cause bit errors and/or complete packet losses, hence, most wireless protocols have automatic repeat requests to increase reliability. The above factors lead to non-deterministic latency and it is necessary for wireless standards to manage the variability accordingly. For example, the IEEE 802.15.4a standard implements guaranteed time slots (GTS) which form a contention free period of a superframe [10]. There are also algorithms, such as greedy perimeter stateless routing 24  (GPSR), being developed so that sensor network packets require the least amount of hops from the source to destination [82]. Finally, wireless standards such as ISA-100.11a and WirelessHart are designed with near real-time industrial communications in mind. 2.5.4  Security  Security is a significant concern for industrial applications since wireless communication is naturally more vulnerable to infiltration than hard-wired systems. However, UWB’s large bandwidth and low power signaling that is embedded near the noise floor inherently make UWB more secure that other wireless systems. The IEEE 802.15.4a has a private ranging mode which requires nodes to exchange a preamble prior to ranging. Although not encrypted, the preamble does prevent impostor attacks and makes eavesdropping more difficult. At 60 GHz, in indoor environments UWB signals have a very short range as they are severely attenuated through wall, hence, security is much less of a concern.  2.6  Conclusions  In recent years, UWB wireless technology in industrial environments has attracted increased attention in both academia and industry. Its potential for precise asset positioning, low power consumption, and resilience to multipath fading gives UWB a competitive advantage over traditional wireless systems for use in industrial environments. In this survey, we have presented a summary of the nature of the industrial networking environment. Wireless technology in industrial environments allows for lower installation and maintenance costs by reducing the need for cables, adds flexibility to plant configurations, and allows for easy asset tracking. These applications, coupled with the success of consumer wireless products, have brought about a strong desire to use wireless technologies in industrial environments. However, the industrial environment poses a significantly different challenge for wireless propagation than the residential and commercial environment. The residential and commercial environment typically has small rooms with many walls and signal fading becomes a problem, whereas, industrial environments are typically characterized by large open areas with a high presence of metallic scatters, hence, time and angular dispersion become significant issues. Using mature wireless market technologies such as 802.11 and 802.15.4, industrial wireless standards such as ISA-100.11a and WirelessHART have been designed to provide reliable 25  and secure wireless operation for non-critical monitoring, alerting, supervisory control, and other control applications. With the emergence of UWB, these applications are further increased. UWB wireless technology enables many industrial applications that had limited performance capabilities, or were previously unachievable, when based on wideband wireless or wired technology. The large bandwidth of UWB technology allows for new types of industrial sensors. The fine time resolution means it is possible to design sensors with sub-meter accuracy for asset tracking. Another industrial UWB sensor application is radar, which in industrial environments can be used for imaging and/or intrusion detection. The low power requirements of UWB technology inherently permit the sensors to be used as part of vast sensor networks since the noise-like properties have negligible interference on existing wireless technologies. Finally, there are also many high data-rate applications possible for short range distribution of multimedia and high-speed data transfer. With the apparent benefits of UWB technology in industrial environments, for successful product development it is necessary to have international standards and accurate channel models. Over the past decade, international regulation and the approval of UWB standards such as ECMA-368 and 802.15.4a has legitimized UWB technology. In turn, this has commenced the development and commercialization of both standards-based and proprietary UWB devices. Accurate channel models will not only help system designers evaluate system reliability and coverage but will also allow them to efficient deploy UWB technology in industrial environments. Measurement-based channel modeling succinctly captures our understanding of the wireless propagation environment in a manner that is useful in the design, test, and simulation of wireless systems. However, as shown, the existing UWB measurement campaigns in industrial environments have been very limited and primarily intended for fair comparison of various PHY and MAC standards. Therefore, the resulting channel models are site-specific and anecdotal. Further research is required to have a complete understanding of the industrial environment and to reliably predict system performance. The aim of this survey was to provide a summary regarding the deployment of UWB technology in industrial environments. We hope that it will be lead to further research of unresolved issues concerning UWB propagation in industrial environments and help users achieve improved system performance. 26  2.7  References  [1]  M. G. diBenedetto, T. Kaiser, A. F. Molisch, I. Oppermann, C. Politano, and D. Porcino, Eds., UWB Communications Systems: A Comprehensive Overview. Cairo, Egypt: Hindawi, 2006.  [2]  M. Z. Win, D. Dardari, A.F. Molisch, W. Wiesbeck and J. Zhang, "History and Applications of UWB," Proc. IEEE , vol. 97, no. 2, pp. 198-204, Feb. 2009.  [3]  R. Scholtz, "Multiple access with time-hopping impulse modulation," in Proc. IEEE Military Commun. Conf. (MILCOM), Boston, MA, vol. 2, pp. 447-450, Oct. 1993.  [4]  M. Z. Win, R. A. Scholtz and M. A. Barnes, "Ultra-wide bandwidth signal propagation for indoor wireless communications," in Proc. IEEE ICC, vol. 1, pp. 5660, Jun. 1997  [5]  M. Z. Win and R. A. Scholtz, "Impulse radio: how it works," IEEE Commun. Letters, vol. 2, no. 2, pp. 36-38, Feb. 1998.  [6]  R. A Scholtz, R.J. Cramer and M. Z. Win "Evaluation of the propagation characteristics of ultra-wideband communication channels," in Proc. IEEE AP-S, vol. 2, pp. 626-630, Jun 1998.  [7]  M. Z. Win and R. A. Scholtz, "Ultra-wide bandwidth time-hopping spread-spectrum impulse radio for wireless multiple-access communications," IEEE Trans. Commun., vol. 48, no. 4, pp. 679-689, Apr. 2000.  [8]  D. Cassioli, M. Z. Win and A. F. Molisch, "The ultra-wide bandwidth indoor channel: from statistical model to simulations," IEEE J. Sel. Commun., vol. 20, no. 6, pp. 1247- 1257, Aug. 2002.  [9]  A. Willig, K. Matheus and A. Wolisz, “Wireless technologies in industrial networks,” Proc. IEEE, vol. 93, no. 6, pp. 1130–1150, Jun. 2005.  [10] A. Willig, "Recent and Emerging Topics in Wireless Industrial Communications: A Selection," IEEE Trans. Ind. Informat., vol. 4, no. 2, pp. 102-124, May 2008. [11] T. S. Rappaport and C. D McGillem, “UHF multipath and propagation measurements in manufacturing environments,” in Proc. IEEE GLOBECOM, pp. 825-831, 1989. [12] T. S. Rappaport and C. D McGillem, “UHF fading in factories,” IEEE J. Sel. Areas Commun., vol. 7, no. 1, pp. 4048, Jan. 1989. [13] T. S. Rappaport, "Indoor radio communications for factories of the future," IEEE Commun. Mag., vol. 27, no. 5, pp.15-24, May 1989. [14] T. S. Rappaport, "Characterization of UHF multipath radio channels in factory buildings," IEEE Trans. Antennas Propag., vol. 37, no. 8, pp. 1058-1069, Aug. 1989. [15] T. S. Rappaport, S. Y. Seidel and K. Takamizawa, “Statistical channel impulse response models for factory an open plan building radio communication system design,” IEEE Trans. Commun., vol. 39, no. 5, pp. 794–807, May 1991. [16] P. Yegani and C.D. McGillem, "A statistical model for the factory radio channel," IEEE Trans. Commun., vol. 39, no. 10, pp. 1445-1454, Oct. 1991. 27  [17] A. Flammini, P. Ferrari, D. Marioli, E. Sisinni and A. Taroni, “Wired and Wireless Sensor Networks for Industrial Applications,” Microelectronics Journal, vol. 40, no. 9, pp. 1322-1336, Sep. 2009. [18] Dick Caro, Wireless Networks for Industrial Automation (3rd Edition). ISA, 2008. [19] V. C. Gungor and G.P. Hancke, “Industrial Wireless Sensor Networks: Challenges, Design Principles, and Technical Approaches”, IEEE Trans. Ind. Electron., vol. 56, no. 10, pp. 4258-4265, Oct. 2009. [20] G.P. Hancke and B. Allen, "Ultrawideband as an Industrial Wireless Solution," IEEE Pervasive Comput., vol. 5, no. 4, pp. 78-85, Dec. 2006. [21] http://www.ieee-ies.org/ [22] http://www.isa.org/ [23] http://www.wina.org/ [24] Federal Communications Commission (FCC), "First Report and Order in The Matter of Revision of Part 15 of the Commission's Rules Regarding Ultra wideband Transmission Systems", ET-Docket 98-153, FCC 02-48, released 22 April 2002. [25] IEEE Standard for Wireless Medium Access Control and Physical Layer Specifications for High Rate Wireless Personal Area Networks (P802.15.3), 2003. [26] ECMA-368, “High Rate Ultra Wideband PHY and MAC standard,” Eur. Comput. Manuf. Assoc., 3rd edition, 2008. [Online]. Available: www.ecma-international.org [27] ECMA-369, “MAC-PHY Interface for ECMA-368,” Eur. Comput. Manuf. Assoc., 3rd edition, 2008. [Online]. Available: www.ecma-international.org [28] P. Fuhr, “Asset Tracking in Industrial Settings—A Review of Wireless Technologies,” Sensors Magazine, Jun. 2008. [29] D. Dardari, C-C Chong and M. Z. Win, "Threshold-Based Time-of-Arrival Estimators in UWB Dense Multipath Channels," IEEE Trans. Comm., vol.56, no. 8, pp. 1366-1378, Aug. 2008. [30] M. R. Mahfouz, A. E. Fathy, M. J. Kuhn and Y. Wang, "Recent trends and advances in UWB positioning," in Proc. IMWS’09, pp.1-4, 2009. [31] IEEE 802.15.4a-2007 standard, “Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (WPANs); Amendment 1: Add Alternate PHYs," Mar. 2007. [32] Z. Sahinoglu and S. Gezici, "Ranging in the IEEE 802.15.4a Standard," Proc. WAMICON '06, pp.1-5, 2006. [33] Decawave, “DW4aSS1000 Advanced Product Information – Scensor”, Datasheet, D0801004DS5, 2009 [34] R. Heile, “PAR for an amendment to existing Standard 802.15.4-2006,” IEEE P802.15-08/0763r1, Jan. 2008. [35] M. McLaughlin, D. Pokrajac, A. Jennings, A. Ward, and T. Harrington, “Merged Proposal for 802.15.4f Active RFID System,” Tech. Rep. Doc., IEEE P802.1509/0802r15 , 2009. 28  [36] R. J. Fontana, L. .A. Foster, B. Fair and D. Wu. “Recent Advances in Ultra Wideband Radar and Ranging Systems”, in Proc. ICUWB, pp. 19-25, 2007. [37] E. Paolini, A. Giorgetti, M. Chiani, R. Minutolo and M. Montanari, “Localization capability of cooperative anti-intruder radar systems,” EURASIP J. on Advances in Signal Processing, Article ID 32082, 14 pages, 2008. [38] Time Domain, “Ultra Wideband Radar Sensor Module”, Datasheet, 2009 [39] R. Knochel, A. Gulck, F. Daschner and O. Schimmer, "UWB-Sensors for Industrial Applications," in Proc. IEEE ICUWB 2007, pp. 529-534, 2007. [40] K-S Lwin, K-J Lim, D-H Shin, N-J Park, D-H Park, K. B. Cho and H-D Kim, “OffLine PD diagnosis for stator winding of Rotating Machines using UWB sensors,” in Proc. CMD, pp.155-158, 2008. [41] S. G. Azevedo and T. E. McEwan, “Micropower impulse radar,” in Science and Technology Review, Livermore, CA: Lawrence Livermore Nat. Lab., (UCRL-52 00096-1/2), pp. 16–29, 1996. [42] F. Althaus, F. Chin, J. Kunisch, J. Pamp, B. Radunovic, U. G. Schuster, F. Trosch, H. L. Truong, M. Weisenhorn, and R. Zetik, “A Comprehensive Investigation of UWB Sensor Networks for Industrial Applications,” in Proc. IST-MWCS, 2006. [43] J. Zhang, “UWB Systems for Wireless Sensor Networks” Proc. IEEE, vol. 97, no 2, pp. 313-331, Feb. 2009. [44] K. S. Low, W. N. Win and M. J. Er, “Wireless Sensor Networks for Industrial Environments” in Proc. CIMCA’05, 2005, pp. 271-276. [45] L. Zheng, “ZigBee Wireless Sensor Network in Industrial Applications,” in Proc. SICE-ICASE, pp.1067-1070, 2006. [46] W. Zeng, H. Wang and A. Xu, “The Research of UWB Ad-Hoc Industrial Network,” in Proc. ChinCom'06, pp. 1-4, 2006. [47] X. Carcelle, T. Dang and C. Devic, “Wireless: Networks in industrial environments: State of the art and Issues,” in Proc. Ad-hoc Networking, pp. 141-156, 2006. [48] P. D. Mitcheson, E. M. Yeatman, G. K. Rao, A. S. Holmes, T. C. Green "Energy Harvesting From Human and Machine Motion for Wireless Electronic Devices," IEEE Proc., vol. 96, no. 9, pp. 1457-1486, Sep. 2008. [49] H. J. Korber, H. Wattar and G. Scholl, “Modular Wireless Real-Time Sensor/Actuator Network for Factory Automation Applications” IEEE Trans. Ind. Informat., vol. 3, no. 2, pp. 111-119, May 2007. [50] P. P. Mercier, D. C. Daly, M. Bhardwaj, D. D. Wentzloff, F. S. Lee and A. P. Chandrakasan, "Ultra-low-power UWB for sensor network applications," in Proc. ISCAS’08, pp. 2562-2565, 18-21 May 2008. [51] A.F. Molisch, P. Orlik, Z. Sahinoglu and J. Zhang, "UWB-based sensor networks and the IEEE 802.15.4a standard - a tutorial," in Proc. ChinaCom'06, pp.1-6, 2006. [52] CEPT - Electronic Communications Committee (ECC), "ECC Decision of 24 Mar. 2006 on the harmonized conditions fordevices using Ultra-wideband (UWB) technology in bands below 10.6 GHz," Doc. ECC/DEC/(06)04, 24 Mar., 2006. 29  [53] Industry Canada, “Devices Using Ultra-Wideband (UWB) Technology”, RSS-220, 2009. [54] A. F. Molisch, "Ultra-Wide-Band Propagation Channels," Proc. IEEE, vol. 97, no. 2, pp.353-371, Feb. 2009. [55] A. F. Molisch, “Ultrawideband Propagation Channels-Theory, Measurement, and Modeling,” IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1528-1545, Sep. 2005. [56] D. Hampicke, A. Richter, A. Schneider, G. Sommerkom, R.S. Thoma and U. Trautwein, "Characterization of the directional mobile radio channel in industrial scenarios, based on wideband propagation measurements," in Proc. IEEE VTC’99, pp. 2258-2262, 1999. [57] S. Kjesbu and T. Brunsvik, "Radiowave propagation in industrial environments," in Proc. IEEE IECON, pp. 2425-2430, 2000. [58] E. Tanghe, W. Joseph, L. Martens, H. Capoen, K. Van Herwegen and W. Vantomme,; "Large-scale fading in industrial environments at wireless communication frequencies," in Proc. AP-S, pp.3001-3004, 2007. [59] E. Tanghe, W. Joseph, L. Verloock, L. Martens, H. Capoen, K. Van Herwegen and W. Vantomme, "The industrial indoor channel: large-scale and temporal fading at 900, 2400, and 5200 MHz," IEEE Trans. Wireless Commun., vol. 7, no. 7, pp. 27402751, Jul. 2008. [60] J. Karedal, S. Wyne, P. Almers, F. Tufvession and A. F. Molisch, "Statistical analysis of the UWB channel in an industrial environment," in Proc. IEEE VTC2004, 2004, pp. 81-85. [61] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A.F. Molisch, "UWB channel measurements in an industrial environment," in Proc. IEEE GLOBECOM, 2004, pp. 3511-3516. [62] M. G. Khan, A. A. Ashraf, J. Karedal, F. Tufvesson and A. F. Molisch, “Measurements and Analysis of UWB Channels in Industrial Environments,” in Proc. IEEE WPMC, 2005. [63] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "A MeasurementBased Statistical Model for Industrial Ultra-Wideband Channels," IEEE Trans. Wireless Commun., vol. 6, no. 8, pp. 3028-3037, Aug. 2007. [64] A.F. Molisch et al., “IEEE 802.15.4a Channel Model Final Report,” Tech. Rep. Doc., IEEE 802.1504-0062-02-004a, 2005. [65] A. F. Molisch, K. Balakrishnan, C. C. Chong, D. Cassioli, S. Emami, A. Fort, J. Karedal, J. Kunisch, H. Schantz, K. Siwiak, and M. Z. Win, “A comprehensive model for ultrawideband propagation channels,” IEEE Trans. Antennas Propag., vol. 54, no. 11, pp. 3151–3166, 2006. [66] J. Kunisch and J. Pamp, "Locally Coherent Ultra-Wideband Radio Channel Model for Sensor Networks in Industrial Environment," in Proc. IEEE ICUWB, 2006, pp.363368.  30  [67] Z. Irahhauten, G. J. M. Janssen, H. Nikookar, A. Yarovoy and L.P. Ligthart, “UWB Channel Measurements and Results for Office and Industrial Environments,” in Proc. IEEE ICUWB, 2006, pp.225-230. [68] M. A. Matin, K. A. Yasmeen, M.A. Mohd Ali, A. K. M. Wahiduzzaman and M. A. Imtiaz, "Statistical model for UWB channel in an industrial environment," in Proc. ICMMT, pp.1015-1017, 2008. [69] K. A. Yasmeen, A. K. M. Wahiduzzaman, A. Imtiaz and M.A. Matin, "Performance analysis of ultra wide band indoor channel," in Proc. MIC-CPE, pp.1-10, 2008. [70] K. A. Yasmeen, M. A. Matin, M. A. Mohd Ali and A. Wahiduzzaman, "Statistical analysis of ultra wideband indoor channel," in Proc. MIC-CPE, pp.1-6, 2008. [71] E. Tanghe, W. Joseph, J. De Bruyne, L. Verloock and L. Martens, "Statistical modelling of power delay profiles and small-scale fading in industrial environments," in Proc. APSURSI, pp.1-4, 2009. [72] E. Tanghe, W. Joseph, J. De Bruyne, L. Verloock and L. Martens, "The Industrial Indoor Channel: Statistical Analysis of the Power Delay Profile,” in Int. J. of Elect. Commun., vol. 64, no. 9, pp. 806-812, Sep. 2010. [73] L. Hentila, V. Hovinen and M. Hamalainen, “Sub-band analysis in UWB radio channel modeling,” in Proc NRS/FWCW, pp. 1–5, 2004. [74] L. Tang, K-C. Wang; Y. Huang and F. Gu, "Channel Characterization and Link Quality Assessment of IEEE 802.15.4-Compliant Radio for Factory Environments," IEEE Trans. on Ind. Informat., vol. 3, no. 2, pp. 99-110, May 2007. [75] L. Clavier, M. Rachdi, M. Fryziel, Y. Delignon, V. Le Thuc, C. Garnier and P.A. Rolland, "Wide band 60 GHz indoor channel: characterization and statistical modeling," in Proc. IEEE VTC’01, pp.2098-2102, 2001. [76] T. Zwick, T.J. Beukem and H. Nam, "Wideband Channel Sounder With Measurements and Model for the 60 GHz Indoor Radio Channel," IEEE Trans. Veh. Technol., vol. 54, no. 4, pp. 1266- 1277, Jul. 2005. [77] C. Park and T.S. Rappaport, "Short-Range Wireless Communications for NextGeneration Networks: UWB, 60 GHz Millimeter-Wave WPAN, And ZigBee," IEEE Wireless Commun., vol. 14, no. 4, pp. 70-78, Aug. 2007. [78] P. Smulders, "Exploiting the 60 GHz band for local wireless multimedia access: prospects and future directions," IEEE Commun. Mag., vol. 40, no. 1, pp. 140-147, Jan. 2002. [79] P. Smulders, "Statistical Characterization of 60-GHz Indoor Radio Channels," IEEE Trans. Antennas Propag. , vol. 57, no. 10, pp. 2820-2829, Oct. 2009. [80] J. Wang; R.V. Prasad and I. Niemegeers, "Analyzing 60 GHz radio links for indoor communications," IEEE Trans. Consum. Electron., vol. 55, no. 4, pp. 1832-1840, Nov. 2009. [81] H. Yang, P. Smulders and M. Herben, “Channel characteristics and transmission performance for various channel configurations at 60 GHz,” in EURASIP J. Wireless Commun. and Networking, vol. 2007, no. 1, pp. 1-15, Jan. 2007. 31  [82] B. Karp and H.T. Kung, “GPSR: Greedy Perimeter Stateless Routing for Wireless Networks,” in Proc. MobiCom 2000, pp. 243-254, Aug. 2000. [83] J. Kolbusz, S. Paszczynski and B.M. Wilamowski, “Network Traffic Model for Industrial Environment,” IEEE Trans. Ind. Informat., vol. 2, no. 4, pp. 213-220, Nov 2006. [84] A. Willig, M. Kubisch, C. Hoene and A. Wolisz, “Measurements of a wireless link in an industrial environment using an IEEE 802.11-compliant physical layer,” IEEE Trans. Ind. Electron., vol. 49, no. 6, pp. 1265–1282, 2002. [85] P. Matkurbanov, S. K. Lee and D. S. Kim, “A Survey and Analysis of Wireless Fieldbus for Industrial Environments” Proc. SICE-ICASE, 2006, pp. 5555-5561. [86] W.C. Chung and D. Ha, "An accurate ultra wideband (UWB) ranging for precision asset location," in Proc. IEEE UWBST, pp. 389-393, 2003. [87] M. Tuchler, V. Schwarz, and A. Huber, "Location accuracy of an UWB localization system in a multi-path environment," Proc. ICU, pp. 414-419, 2005. [88] V. Schwarz, A. Huber and M. Tuchler, "Accuracy of a commercial UWB 3D location/tracking system and its impact on LT application scenarios," Proc. ICU, 2005, pp. 599-603. [89] D. Dardari, A. Conti, U. Ferner, A. Giorgetti and M.Z. Win,, "Ranging With Ultrawide Bandwidth Signals in Multipath Environments," IEEE Proc., vol. 97, no. 2, pp. 404-426, Feb. 2009. [90] B. J. Harker, "Radar Level-Sensing Instrument for Industrial (and Automotive) Applications," in Proc. EUMC’07, pp.1-4, 2001. [91] W. Zeng, H. Wang, H Yu and A. Xu, “The Research and Application of UWB Based Industrial Network,” in Proc. UWBUSIS, pp. 153-155, 2006. [92] J. Foerster et al., “Channel Modeling Subcommittee Final Report,” IEEE, Document IEEE 02490r0P802-15 SG3a, 2003. [93] J. Ellis, P. Kinney and J. Lampe, “IEEE P802.15.4a WPAN Alternate PHY – PAR,” IEEE P802.15-04/048r1, Jan. 2004. [94] IEEE 802.15.4 standard, “Part 15.4: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for Low-Rate Wireless Personal Area Networks (LR-WPANs)”, Oct. 2003. [95] IEC/PAS 62591 standard, “Industrial communication networks - Fieldbus specifications – WirelessHARTTM communication network and communication profile”, 1st edition, 2009. [96] ISA 100.11a standard, “Wireless Systems for Industrial Automation: Process Control and Related Applications” 1st edition, 2009. [97] ECMA-387, “High Rate 60 GHz PHY, MAC and HDMI PAL,” Eur. Comput. Manuf. Assoc., 1st edition, 2008. [Online]. Available: www.ecma-international.org  32  [98] IEEE 802.15.3c-2009 standard “Part 15.3: Wireless Medium Access Control (MAC) and Physical Layer (PHY) Specifications for High Rate Wireless Personal Area Networks (WPANs) Amendment 2: Millimeter-wave-based Alternative Physical Layer Extension,” Oct. 2008. [99] A. Miaoudakis, A.Lekkas, G.Kalivas and S. Koubias, "Radio channel characterization in industrial environments and spread spectrum modem performance," in Proc. ETFA’05, pp. 87-93, 2005. [100] A. F. Molisch, J. R. Foerster and M. Pendergrass, "Channel models for ultrawideband personal area networks," IEEE Trans. Wireless Commun., vol. 10, no. 6, pp. 14-21, Dec. 2003. [101] B. Gaffney, “White Paper: Combined LOS and NLOS UWB Channel Model” Decawave Doc. D0712003WP, Jan. 2008. Available: http://www.decawave.com/ [102] G. Llano, J. Reig, L. Rubio and A. Garcia, “Ultra-wideband Frequency Analysis: State-of-the-art, Measurements and Modeling,” WAVES, vol. 1, pp. 147-154, 2009. [103] H. Gong; H. Hong and Z. Chen, "Performance of UWB Systems with Suboptimal Receivers under IEEE 802.15.4a Industrial Environments," in Proc. CNSR, pp.283286, 2007. [104] T. Gigl, J. Preishuber-Pfluegl and K. Witrisal, "Statistical analysis of a UWB energy detector for ranging in IEEE 802.15.4a," in Proc. ICUWB, pp.129-134, 2009. [105] E. Tanghe, W. Joseph, L. Verloock, L. Martens, H. Capoen, K. Van Herwegen and T. Buysschaert, "Range of an IEEE 802.11b/g system in industrial environments based on site survey measurements and propagation models," in Proc. AP-S, pp.1-4, 2008. [106] E. Tanghe, W. Joseph, L. Verloock, L. Martens, H. Capoen, K. Van Herwegen, T. Buysschaert, "Statistical validation of WLAN range calculated with propagation models for industrial environments by chipset-level received signal strength measurements," Science, Measurement & Technology, IET, vol.3, no. 3, pp.244-255, May 2009.  33  Chapter 3 Computer-Assisted Identification of Clusters within UWB Channel Impulse Responses 3 3.1  Introduction  Past work by various groups, and most notably the IEEE 802.15.3a and 802.15.4a channel modeling committees, has contributed greatly to our understanding of the nature and structure of ultra-wideband (UWB) channel impulse responses (CIRs) in residential, commercial, outdoor and industrial environments. The resulting models, however, are intended only for fair comparison of physical and media-access control layers. Hence, there exists an ongoing need to analyze and model the detailed structure of UWB CIRs observed in both specific sites and new environments not considered previously as the detailed structure of the CIR plays a particularly significant role in the design of signaling schemes and receiver architectures [1]. The tendency for multipath components (MPCs) in wideband impulse responses to form clusters was first reported by Saleh and Valenzuela over twenty years ago [2]. As the bandwidth of the channel frequency response measurement increases, the temporal resolution of the corresponding channel impulse response also increases. For wider bandwidths, the tendency for multipath components to cluster in time, often with a characteristic exponential decay profile, was often observed. They modeled this phenomenon as follows: (1) the individual components have independent uniform phases and independent Rayleigh amplitudes that decayed exponentially with cluster and ray delays and (2) the cluster and ray arrival times have independent, but fixed, Poisson processes. This mathematical description of channel impulse response in wideband channel later became known as the S-V Model. A modified S-V model was later adopted by both the IEEE 802.15.3a and 802.15.4a wireless personal area networks task groups as the basis for standardized models of the UWB CIRs that one is likely to encounter in residential, office, outdoor and industrial  A version of this chapter will be submitted for publication: A. Stefanski, J. Chuang, S. Bashir and D.G. Michelson, “Computer-Assisted Identification of Clusters within UWB Channel Impulse Responses.”  34  environments [3]-[5]. As MPCs observed in certain dense scattering environments are often so closely spaced in time that clustering can no longer be observed in the CIR, IEEE 802.15.4a also specified a dense single-cluster model to be used in cases such as NLOS channels in office and industrial environments where the sparse multi-cluster S-V model is not appropriate [6]-[7]. Despite the tremendous effort that has been applied to UWB channel measurement and modeling in recent years (and the importance of cluster identification to the process), most researchers still identify clusters in CIRs through relatively time-consuming ‘visual identification’ techniques that rely on subjective assessment by the analyst. A reliable UWB cluster identification algorithm has the potential to make cluster identification less subjective, more consistent, and less time consuming. This would allow researchers to process larger data sets more quickly and reliably, and thereby, allow them to draw broader conclusions. However, even with the apparent benefits, only a few researchers have attempted to develop such an algorithm [8]–[11]. The cluster-based S-V model is very compelling because it often appears to provide an acceptable fit to measured data, however, three key factors have likely contributed to the lack of progress to date. Foremost, although it is generally accepted that clusters are groups of MPCs that have similar large-scale properties such as time-of-arrival, angle-of-departure, and angle-of-arrival, there is a surprising lack of agreement concerning the definition of a cluster. For example, while most have assumed that a cluster has finite length, 802.11 TgN assumes that once a cluster begins, it continues indefinitely [12]. Secondly, as noted in [5] and [13], there is surprisingly a lack of agreement concerning the criteria for distinguishing clusters. Thirdly, as noted in [8] and [14], there is a perception that automated clustering techniques are not sufficiently robust for practical use. In this chapter, we propose an alternative method for rigorous computer-assisted cluster identification that improves upon past work. [8] proposed a clustering algorithm for measured channel impulse responses, however, the authors acknowledge the algorithm fails to distinguish overlapping clusters which readily occur in dense environments. More recently, [9] and [10] have proposed improved methods which searching for local maxima and cluster delays, however, the selected cluster start-time indices may not necessarily provide the best-fit to the measured data. [11] proposes an identification method which first considers angle-of-arrival and, in a second step, time-of-arrival but once again does not 35  validate the fit. To date, all of the existing methods attempt to replicate the ‘visual inspection’ process, however, our proposed algorithm uses a least-squares fit method to identify clusters, as suggested in [5]. We build upon the work presented in [15], which uses a simple least-squares linear regression to identify the clusters, by utilizing multivariate adaptive regression spline concepts to develop a more robust algorithm that is less sensitive to noise and outliers. We focus on the manner in which clusters of MPCs of a given start time, duration, and exponential decay introduce discontinuities in the shape of the CIR and have formulated the algorithm in a MATLAB application such that it can be used a basis for interactive aid to the analyst. Furthermore, we have validated and verified our algorithm using both: (a) simulated 802.15.4a channel model parameters and (b) measured channel impulse response collected in line-of-sight (LOS) office and industrial environments. The remainder of this chapter is organized as follows: In Section 3.2, we briefly review the UWB standard channel models adopted by the IEEE 802.15.4a task group and summarize our approach to cluster identification. In Section 3.3, we describe the steps that comprise our cluster identification algorithm. In Section 3.4, we describe the role of the analyst and introduce penalties to enforce fundamental S-V model assumptions. Section 3.5 present the results of our efforts to validate the algorithm using CIRs generated using 802.15.4a simulation code and measured CIRs from both LOS office and industrial environments. In Section 3.6, we consider possible sources of error that could affect the performance of the algorithm. Finally, in Section 3.7, we summarize our contributions.  3.2  Cluster Identification Approach  Our approach to cluster identification is based on identification of amplitude discontinuities in the shape of the CIR. The IEEE 802.15.4a task group has recommended using either a sparse multi-cluster or a dense single-cluster model, as appropriate, to represent UWB CIRs [4]. The sparse multi-cluster model is based upon the well known SV model. The general shape of the CIR, which does not include the distortion of the MPCs by the frequency dependence of path loss, is given by L  Kl  h ( t ) = ∑∑ ak ,l exp ( jφk ,l ) δ ( t − Τl − τ k ,l ) ,  (3.1)  l =0 k =0  where the MPCs are modeled as Dirac delta functions, δ(.), and ak,l and φk,l are the amplitude and phase of the kth MPC in the lth cluster. L is the total number of clusters in 36  the CIR and Kl is the total number of MPCs within the lth cluster. Tl and τk,l represent the arrival time of the lth cluster and the kth MPC in the lth cluster, respectively [4]. The envelope of the CIR is defined as the square magnitude of its components and it describes the manner in which the power in the CIR decays with delay. Although the usage is not strictly correct in the context of UWB because the wide sense stationary uncorrelated scattering (WSSUS) assumption no longer holds, we shall follow the common practice of referring to it as the power delay profile or PDP. When small-scale fading is suppressed, the shape of the PDP is described by the product of two exponential functions,  { } ∝ exp ( −Τ / Γ ) exp ( −τ  E ak ,l  2  l  k ,l  /γ ) ,  (3.2)  where Γ and γ are the cluster and intracluster decay constants respectively, as depicted in Figure 5(a). The arrival time of the clusters and MPCs, Tl and τk,l, are found to follow the Poisson distribution with arrival rates, Λ and λ, respectively. For the dense single-cluster model, i.e., without any discontinuities, the shape of the PDP takes the form,  { } ∝ 1− χ exp  −γτ  E ak ,l  2  k ,l  rise    −τ k ,l    exp   , γ  1    (3.3)  where χ denotes the attenuation of the first component, γrise determines how fast the PDP rises to its local maximum, and γ1 represents the decay at later times. The arrival rate of the MPCs is fixed, ∆t, given by the inverse of the signal bandwidth, as depicted in Figure 5(b). In both models, the small-scale fading follows the Nakagami distribution. The shape of a cluster may be greatly distorted by the effects of noise and small-scale fading. As a result, our observation of the PDP, when expressed on a decibel scale, takes the form  ydB (τ ) = fdB (τ ) + ε dB (τ ) ,  (3.4)  where fdB(τ) is the actual PDP and takes the form of either (2) or (3) as appropriate and εdB (τ) is a random variable that accounts for the combined effects of white noise and smallscale fading. Because we express the envelope on a semi-logarithmic scale, exponential decay appears as a straight line. Using linear regression techniques, we find a set of nonoverlapping piecewise linear regression lines, gdB(t), that approximates our observations of the PDP expressed on a decibel scale, i.e., 37   b10 + b11τ b + b τ  20 21 f dB (τ ) ≈ g dB (τ ) =  M  b ˆ + b ˆ τ L1  L0  t1 ≤ τ < t 2 t 2 ≤ τ < t3 M t Lˆ ≤ τ  ,  (3.5)  where bi0 and bi1 are the intercepts and slopes of the best fit lines, respectively, and ti is the estimated arrival times for each of the estimated L̂ clusters. Each regression line in gdB(τ) corresponds to a cluster and the estimated arrival times, ti, correspond to the discontinuities in the PDP. Once gdB(τ) has been estimated, ti will provide us with the accurate start of the clusters. Then, from the estimated arrival time of each cluster, the parameters of the S-V model can be extracted as outlined in [16].  Figure 5: Parameters of the (a) sparse multi-cluster (S-V) and (b) dense single-cluster channel impulse response models.  38  3.3  Description of the Cluster Identification Algorithm  The goal of our cluster identification algorithm is to determine the combination of exponentially decaying clusters that best fit the most significant MPCs in the power delay profile. Whether measured in the time or frequency domain, a measured channel response has a finite bandwidth that is determined by the instrument and/or the measurement process. The result is equivalent to convolving the true CIR with a sinc function whose duration is inversely proportional to the bandwidth of the measurement. Before processing a measured CIR using our cluster identification algorithm, one must first remove the effects of the finite bandwidth either by windowing or deconvolution. Other pre-processing steps that must be performed include setting thresholds for noise, suppressing small-scale fading, and so forth, as outlined in [16]. As the first set of results presented in Section 3.5 are based upon simulated PDPs, we were able to omit this pre-processing step. The measured results used for validation have been adequately pre-processed. Computing the spatial average of several PDPs over a local area is one method for suppressing small-scale fading, and it should be applied whenever possible to obtain a better estimate of the shape of the PDP. Because the first set of results is based upon randomly simulated PDPs, spatial averaging could not be applied. Accordingly, we developed an alternative method for small-scale fading suppression that involves using an averaging filter in the time domain to smooth the rapid variations within the CIR caused by small-scale fading. This approach has been used in image processing to reveal edges against background noise [21]. Using such techniques allows one to apply our cluster identification algorithm to either spatially-averaged or instantaneous PDPs. A flow chart of our cluster identification algorithm is depicted in Figure 6. It includes the three main steps that we describe in more detail below: (1) local smoothing in the time domain, (2) the initial search for clusters by identifying gaps in the PDP, and (3) the iterative search for clusters based upon trial fitting of intracluster decay curves to the PDP. We have also formulated the algorithm in a MATLAB application such that it can be used a basis for interactive aid to the analyst, the details of which are provided in Appendix A.  39  Figure 6: Flowchart of the rule-based cluster identification algorithm 40  3.3.1  Local Smoothing in the Time Domain  The first step is to suppress small-scale fading through local smoothing in the time domain. The PDP will typically be sampled at an interval, ts, that is inversely proportional to the bandwidth of the measured signal. In the UWB case, the bandwidth will typically range from a minimum of 500 MHz to a maximum of 7.5 GHz yielding 0.13 ns < ts < 2 ns. We smooth the PDP in a two-step process. First, we apply a running average filter defined by  PDPs [ m] =  1 N /2 ∑ PDP[ m + i] , M i =− N /2  (3.6)  where N is the user-defined smoothing interval and M will nominally be equal to N + 1. Next, the averaged PDP is decimated by a factor, k, where  PDPs [ k ] = PDPs [ km] .  (3.7)  Decimating the time series reduces the chance of having outliers in the linear regression analysis applied later. While local smoothing is an essential step, the analyst should be aware that it may distort steeply decreasing clusters. In certain cases, a given tap may not contain any energy after we apply an amplitude threshold. In such cases, that tap should be eliminated and the value of M should be reduced to properly reflect the number of energybearing taps in the PDP. A representative PDP that has been generated using the standard channel impulse response generator developed by the IEEE 802.15.4a channel modeling committee is given in Figure 7. The dots represent the components of the PDP that result when the averaging filter given in (6) is applied with N = 9. 3.3.2  Initial Search for Clusters  If the interval between successive MPCs exceeds a user-defined threshold, we consider this to be a gap between distinct clusters. The number of such gaps plus 1 gives the initial number of cluster Linit. Such gaps often arise in simulated PDPs but, in our experience, rarely arise in measured data. The user can select an appropriate value for this threshold by inspecting the PDP.  41  -20  PDP [dB]  -40  -60  -80  -100  0  50  100  150 200 Time [ns]  250  300  Figure 7: A locally smoothed power delay profile  3.3.3  Iterative Search for Clusters  Our cluster identification algorithm is based upon expression of the MPCs on a semilogarithmic scale so that exponential intracluster decay profiles will be displayed as straight lines with given slopes. We use piecewise linear regression methods to determine the particular combination of straight lines that best fits contiguous groups of significant MPCs, given by Eqn. (3.6) in a least squares sense. The search begins with the assumption that the initial number of clusters is  L̂ = L in it  as  determined by the methods described in Section 3.3. The algorithm then estimates goodness-of-fit by calculating the RMS error over the entire PDP, RMSE =  1 M  M  ∑ PDP [ m ] − g [ m ] s  2  ,  (3.8)  m =1  where M is the total number of smoothed MPCs identified and g[m] is g(τ) sampled at intervals Nts. If the resulting RMS error of any cluster exceeds a specified threshold, it is assumed that an unidentified cluster is causing a discontinuity in the PDP. In response, the algorithm increments the number of clusters by 1, tries all possible combinations of L̂ straight lines that one might fit to contiguous groups of MPCs to form L̂ clusters, then selects the combination that yields the best fit, i.e., the lowest RMS error. If the best fit 42  satisfies the error threshold criterion, the process ends, and the number of clusters in the CIR is reported as L̂ . Otherwise, it is assumed that the number of clusters must be incremented further and the process is repeated. A typical sequence is depicted in Figure 8. 3.3.4  Recursive Partitioning  An exhaustive search of all possible cluster configurations will ensure that the best-fit solution is ultimately found. However, the number of possible combinations increases rapidly as the number of clusters increases and soon becomes computationally intractable. During trials conducted upon both simulated and measured CIRs, we have observed that incrementing the number of clusters used to represent a CIR typically involves subdividing an existing cluster. In statistics, this approach is referred to as recursive partitioning [18]. Here, the decision to subdivide an existing cluster is also based upon the RMS error being further reduced. Making this a rule in our search strategy and eliminating all other possibilities reduces the number of trials that we must conduct from Ο(ML-1) to just Ο(M) where M is the total number of smoothed MPCs in the PDP and L is the number of clusters in the CIR. Implementing such a partitioning strategy dramatically reduces the number of combinations that must be examined compared to an exhaustive search. This partitioning strategy plays an important role in making our algorithm tractable.  43  Figure 8: Evolution of the cluster identification process: (a) two clusters are broken into (b) three clusters, the second of which is broken on the next iteration to yield (c) four clusters. 44  3.4  Role of the Analyst  There are difficulties in developing a completely autonomous algorithm because of noise and small-scale fading seen in UWB channel measurements, however, as mentioned before the visual identification of clusters is not objective nor rigorous. We are able to overcome these two problems by using the algorithm in Section 3.3 with the application of lack-of-fit penalties by the analyst. The use of penalties makes the algorithm tractable and aids in keeping the analyst objective. Rather than simply adjusting the start time of the cluster, the analyst is required to weigh additional rules to enforce the shape of the modified S-V model. Through the use of a lack-of-fit criterion, when an additional rule is violated, the increase in the LOF criterion will reduce the likelihood that a cluster will be subdivided at the violating location. The rules, which are explained in detail in the following sections are not defined in advance (i.e. βx = 1 by default) and are applied interactively by the analyst. 3.4.1  Generalized Cross-Validation Criterion  As mentioned in Section 3.3.3, a RMS error threshold sets the stopping condition for the algorithm. However, in order to ensure that the fast variations of the PDP are not interpreted as clusters, we make the stopping condition more robust by adopting the generalized cross-validation (GCV) criterion,  GCV =  RMSE ,  αB 1 − M   (3.9)  that has been used in [18] and elsewhere. Here, B is the number of parameters that we seek to estimate (in our case, the number of slopes and intercepts that define g(τ)), M is the number of smoothed MPCs and α is the user-defined penalty coefficient. As the number of identified clusters increases, the GCV will initially decrease with the RMS error since new clusters will significantly reduce the error. Past a point defined by the penalty coefficient, α, the increasing number of parameters B will cause the GCV to increase as further error reduction will be small. Based on the GCV criterion, the algorithm will stop: (1) when the value of GCV is below the user-defined threshold or (2) when the penalty prevents the value of GCV from being reduced further.  45  3.4.2  Additional Penalties  In order to ensure the correct operation of the cluster identification algorithm, we enforce three additional rules. The three additional rules are jointly enforced during recursive partitioning by applying a multiplicative penalty coefficient, βi, where i identifies the additional rule. The decision to split an existing cluster is now based on the lack-of-fit (LOF) criterion,  LOF = β1β2 β3GCV = β1β2 β3  RMSE .  αB 1 − M   (3.10)  The increase in the LOF criterion that occurs when an additional rule is violated will reduce the likelihood that a cluster will be subdivided at the selected location and the algorithm will continue to search for a more optimal breakpoint. However, it does not completely stop the algorithm from identifying the location as a cluster if this sufficiently reduces the RMS error compared to other candidate locations. 3.4.2.1 Minimum Length Penalty – β1  The cluster lengths must exceed a minimum number of MPCs otherwise the β1 penalty is applied. A typical scenario where the β1 penalty has to be enforced is for LOS measurements when the LOS peak is split between 1-2 delay bins. 3.4.2.2 Excess Power Level Penalty – β2  In the modified S-V model, the intercluster decay rate (ie: Γ) is typically greater than the intraculster decay rate (ie: γ), hence, the power level at the start of a given cluster must be greater than the residual power level extrapolated from the previous cluster. If at the algorithm-selected breakpoint, the extrapolated power is within a user-defined power level the β2 penalty is applied. 3.4.2.3 Positive Slope Penalty – β3  The S-V model assumes that the clusters decay, hence, the slopes of the regression lines that we fit to the MPCs, must be negative because. At times, the algorithm may find a positive sloped cluster because of noise and/or a very slow onset caused by dense clustering. The latter is treated as a special case in the 802.15.4a channel model. 46  3.5  Validation  3.5.1  Simulated 802.15.4a CIRs  The performance of the algorithm was first validated using simulated CIRs generated by the IEEE802.15.4a simulator. This allowed for objective validation, since the exact number of clusters and start time of each cluster in every CIR could be extracted from the simulator and compared against the number and start times of clusters identified by our algorithm. A total of 60 CIRs were generated. 15 CIRs each from CM3, CM5, CM6 and CM7 were taken in order to have a fair representation of the various environments that exhibit varying degrees of clustering. The distribution of CIRs with respect to number of clusters is given in Figure 9. Each CIR was converted into PDP in following the process described in [5].  Figure 9: Distribution of Clusters Among Simulated CIR Each CIR was then analyzed using the graphical user interface of the algorithm. The RMS error threshold was adjusted to the extent where no more or less clusters could be identified correctly by either reducing or increasing the threshold. Since the simulated CIRs were assumed to follow the modified S-V model [5], the penalty coefficients α, β1, β2 and β3 were set to one for most cases. β1, β2 and β3 were set to 2 if a cluster with positive slope was identified. Overall, 68% of the CIRs resulted in 80% or more clusters being identified correctly. The remaining 32% CIRs resulted in 70% or less clusters being identified correctly. Figure 10 shows the distribution of estimation accuracy of the algorithm. 47  Figure 10: Estimation Accuracy using Simulated CIRs The effectiveness of the algorithm generally decreased as the number of clusters in a given CIR increased. It was observed that the algorithm can most effectively identify clusters when the number of clusters is between 2 and 5. The estimation accuracy decreases to below 70% when the CIRs contain large number of clusters, i.e. 15 or more, as shown in Figure 11.  Figure 11: Effectiveness of Algorithm for Simulated CIRs For most of the CIRs processed, the effective range of minimum RMS error threshold was found to be between 1.8dB to 3dB. In 42% of the CIRs, all or most of the clusters were 48  found with RMS error threshold between 1.8dB and 2.2dB; in another 25% of CIRs, most clusters were found between 2.3dB and 3dB. Figure 12 shows the distribution of RMS error thresholds that gave the most accurate results in all of the CIRs processed. It was observed that the CIRs with large number of clusters (15 or more) required minimum RMS error threshold of 3dB or more; reducing the threshold below this level did not result in any improvement in the number of clusters correctly identified. Hence an appropriate choice of the threshold level is one, below which the desired clusters start breaking into smaller ones, or above which the correctly identified clusters start merging to form larger ones.  Figure 12: RMS Error Threshold for Correct Cluster Identification Some of the CIRs generated by the IEEE802.15.4a simulator contained clusters that were too dense and hence our algorithm could not locate them. However, it is worth noting that these clusters could not be identified visually either. About 20% of the simulated CIRs contained at least one cluster that fell into this category. An example CIR is shown in Figure 13a. Moreover, some CIRs contained clusters that did not appear to strictly follow the underlying model assumptions, particularly the assumption of exponentially decaying power of MPCs within a cluster [5]. Although the cluster was generated with a negative decay constant, however, the addition of small-scale fading made the cluster appear positive. Our algorithm was nevertheless able to detect the start times of many such clusters. One such CIR is shown in Figure 13b. 49  Figure 13: Example of CIRs where (a) clusters #2, #4, and #12 could not be identified and (b) appears to have a positive slope  50  3.5.2  Measured UWB CIRs  3.5.2.1 Data Acquisition and Processing  In order to acquire measured CIRs, we performed frequency-domain channel sounding measurements following the recommendations of [1] and [5]. In-depth descriptions of the UWB channel sounder setup and measurement campaigns and also the detailed MATLAB code are available in Appendix B and C, respectively. Our UWB channel sounder consisted of an Agilent E8362B network analyzer which measured the frequency response by sweeping from 3.1 GHz to 10.6 GHz in 6401 linearly distributed steps. Two omni-directional UWB biconical antennas were used, one of which was mounted on a tripod and the other on a 2-D linear positioner. At each measurement location, the transmit antenna was kept in a fixed position and the receive antenna was moved around a square 7x7 virtual array with λ/2 spacing (48.35 mm at 3.1 GHz) using the positioner. Both antennas were mounted 1.25m above the ground. The UWB measurements were conducted at the University of British Columbia inside the MacLeod Building and the Advanced Materials and Process Engineering Laboratory (AMPEL). The MacLeod Building is representative of a typical indoor office environment and has many hallways, classrooms, and laboratories. The walls are made of cinderblocks. AMPEL is representative of a large indoor industrial environment as it includes a 5 tonne gantry crane, a 5.5m long rotary kiln, fluidized bed reactors, 20m long runout table, and other metallurgical processing facilities. The floor area measures 1500m2, half of which has a ceiling height of 8m with metal trusses and the other half has a 3m concrete ceiling. Our measurement database included 40 LOS locations in various rooms inside the MacLeod Building and 45 LOS measurements inside AMPEL, corresponding to CM3 and CM7 scenarios respectively. The measurement distance was varied logarithmically between 1m and 18m and we captured 4165 channel responses in total. The subsequent post-processing steps were applied to compensate for the non-linear effects of the instrumentation and the antenna patterns. A through-line calibration removes the losses due to the cables and amplifiers, however, the task of de-embedding the antennas is far more difficult as outlined in [1] and [19]. To completely de-embed the propagation channel response from the radio channel it is necessary to know the radiation pattern of both antennas and then measure the frequency-dependant double-directional response to 51  determine the angle-of-departure and the angle-of-arrival using advanced algorithms such as CLEAN or SAGE. However, these algorithms are very complex and, typically, the process is simplified by assuming an antenna transfer function that is averaged in all directional or only in the azimuthal plane. Similar to [14], for the purpose of this chapter we measured the radiation pattern of both antennas in an anechoic chamber and averaged the respective antenna transfer functions in all directions. Following [7] and [14], a Hamming windows was applied to suppress energy between delay bins and the frequency responses were transformed into instantaneous channel impulse responses, h(τ), using the real passband processing technique. In order to remove small-scale fading effects, the averaged power delay profile (APDP) was obtained by summing the magnitudes of h(τ) across the virtual array, as defined by  APDP(τ ) =  1 M N 2 h(τ , m, n) ∑∑ MN m=1 n=1  ,  (3.11)  where M and N are the number of virtual elements (ie: 7). Finally, the maximum peak of the APDP was set to 0dB and the initial time delay was removed. 3.5.2.2 Power Delay Profile Analysis The visual inspection of our measured APDPs revealed some important observations. All of the measured responses are continuous in power and large discontinuities are not evident. The responses were more similar in shape to [7] and [14] than 802.15.4a simulated responses. All responses exhibit a strong LOS peak which on average contains 17% of the total energy. Furthermore, the energy in the LOS peak decreases exponentially with distance. We also found the delay spread to increase with distance and to be larger industrial environments than office environments. In our measured APDPs, we found on average the existence of 3–4 clusters in CM3 and 1–3 clusters in CM7, as summarized in Table I. This corresponds well with other measurement campaigns which have typically report the number of clusters to range from 1 to 5 [20]. The time of arrival of the second cluster is on average 65ns, for both CM3 and CM7, and appears to decrease with delay. Considering the size of our measurement environment, this could mean that clusters are more likely to be caused by multiple bounces rather than single bounces. Two CIRs and APDPs that are representative of those observed for CM3 and CM7 are shown in Figure 14 and 15, respectively. 52  TABLE 6: NUMBER OF CLUSTERS OBSERVED IN MEASURED RESPONSES CM3 (Office LOS) Total Number of PDPs % PDPs w/ 1 cluster % PDPs w/ 2 clusters % PDPs w/ 3 clusters % PDPs w/ 4 clusters % PDPs w/ 5 clusters % PDPs w/ 6 clusters % PDPs w/ 7 or more clusters  40 0% 0% 33 % 35 % 18 % 10 % 6%  CM7 (Industrial LOS) 45 29 % 38 % 20 % 7% 4% 2% 0%  Figure 14: Representative Plots of CIRs and APDPs for CM3  53  Figure 15: Representative Plots of CIRs and APDPs for CM7 Next, we proceeded to identify clusters using our computer-assisted cluster identification algorithm. On average, we identified 3.3 clusters in the office environment and 2.8 clusters in the industrial environment. For the identification process, we typically used threshold of -40dB and a delay of 160ns which corresponds to approximately 50m. We used all of the MPCs for the identification process since the APDPs all had the small-signal fluctuations removed. For penalties, we typically had to set β1 = 1.5 (+/- 0.25) to remove the short but strong cluster found due the LOS component. Occasionally, we set β2 = 1.2 (+/- 0.1) to remove false clusters found due to small noise fluctuations. We set β3 = 2 in the instance a positive-sloped cluster was identified. Finally, we extracted the S-V parameters for our measured data in order to validate that we correctly identified the clusters. Our estimates for the model parameters are summarized in Table 7. Comparing to the 802.15.4a model, the cluster arrival rate (1/Λ) and cluster power decay (Γ) corresponded closely. However, we typically identified 2 fewer clusters because dense clusters could not be split reliably. For the ray power decay rate, we have found that it is not constant and that in both CM3 and CM7 γ increase with delay. Lastly, the concept of the ray arrival rate, λ, loses meaning since the responses are continuous in power over time.  54  TABLE 7: EXTRACTED S-V PARAMETERS FOR MEASURED UWB RESPONSES  Lmean 1/Λ [ns] Γ [ns] γ [ns] a  3.6  CM3 (Office LOS) 802.15.4a Measured Model 5.4 3.3 62.5 40.9 14.6 12.8 6.4 – -  CM7 (Industrial LOS) 802.15.4a Model Measured 4.75 14.1 13.47 0.651 0.926  2.8 30.1 17.8 -  Sources of Error  Several factors may affect the performance of the algorithm including: (1) noise and small-scale fading, (2) overlap between clusters, (3) the presence of anomalous clusters and (4) coupling between the input parameters. We discuss each of these in more detail below. 3.6.1  Noise and Small-Scale Fading  The LMS linear regression techniques that we use to fit trial PDPs to groups of MPCs can be sensitive to extreme values or outliers that affect the slope and offset of the resulting regression lines. In UWB propagation measurements, small-scale fading is always present and can significantly mask the shape of the PDP [4]. For example, a PDP whose features are buried by small-scale fading is depicted in Figure 16.  Figure 16: A PDP whose features have been buried by small-scale fading. 55  To illustrate the effect of small-scale fading on cluster detection, assume that there are L clusters in the PDP and that we have only fit L̂ regression lines. When L̂ < L , the residual error can be expressed as the sum of two components,  ε (τ ) = bias (τ ) + Xσ ,  (3.12)  as depicted in Figure 16. The random variable, Xσ, is the combination of small-scale fading and measurement noise and the bias term is the difference between the envelope of the PDP and the current best regression line, g(τ). The bias term in decibel scale can be described as,  {  biasdB (τ ) = 10 log10 E a (τ )  2  } − g (τ ) ,  (3.13)  where E{|a(τ)|2} is the envelope of the PDP as described by (2) and (3). Because the random error cannot be eliminated, the success of the algorithm rests on whether or not we can successfully reduce the RMS error by reducing the bias term with a better estimate, i.e., generate a more accurate g(τ). For example, the error in the first regression line in Figure 17 can be reduced while the error in the second regression line cannot. As a result, cluster detection will be difficult when small-scale fading is significant.  Figure 17: The components of the residual error 56  3.6.2  Overlap between Clusters  The identification of clusters is also directly related to how the clusters arrive. In the extreme case, when the arrivals of all clusters are too close in time, the clusters become indistinguishable based upon the shape, e.g., the dense single-cluster model used in dense scattering environments. Figure 18 shows an example of the identified clusters in a channel realization from CM5, the outdoor LOS environment. The lines represent the identified clusters and the crosses indicate where the actual clusters start. While most clusters are easily resolved, two clusters around 190 ns and two more around 250 ns cannot because they strongly overlap. 0  PDP [dB]  -20  -40  -60  -80  -100 0  50  100  150 200 Time [ns]  250  300  350  Figure 18: The ten identified clusters in this CM5 CIR are denoted by bold lines. The starts of the twelve actual clusters are marked by crosses To a lesser extent, the cluster decay rate Γ and the MPC decay rate γ also affects our ability to distinguish the start of a new cluster. If the first cluster has a continuous decay profile, the arrival of a second cluster ∆τ after the start of the first inserts a discontinuity of height ∆ PDP (τ 1 ) =  exp ( − ∆τ Γ ) , exp ( − ∆τ γ )  (3.14) 57  or, on a decibel scale, as 1 1 ∆ PDP, dB (τ 1 ) = k ∆τ  −  , γ Γ  (3.15)  where τ1 is the arrival time of the new cluster and k is 10log10(e). This discontinuity is essentially the bias term in (3.12). As the MPC decay rate increases, the clusters can arrive increasingly closer in time and still be successfully distinguished. From (3.14), we calculated the average magnitude of the discontinuities in the PDP for the standard channel models assuming that, on average, ∆τ is the inverse of the cluster arrival rate, Λ. All channel models give reasonably distinct clusters except for CM2 – Residential NLOS where the discontinuities, on average, are only 0.7 dB. This makes it difficult to distinguish clusters in CM2 whether by analyst or by algorithm and suggests that it might be more appropriately represented by a single-cluster model. Inspection of typical CM2 CIRs tends to support this notion.  3.7  Conclusions  We have developed an automated cluster identification algorithm that determines how a UWB channel impulse response (CIR) can be most effectively represented by either of the IEEE 802.15.4a standard channel impulse response models: (1) a single exponentially decaying cluster (a straight line when expressed on a semi-log scale) or (2) a sequence of exponentially decaying clusters described by the Saleh-Valenzuela model, as appropriate. Trials conducted using UWB CIRs generated by a simulation code developed by IEEE 802.15.4a and UWB CIRs measured in office and industrial environments have confirmed the validity of our approach. Although the algorithm works best when applied to CIRs that have been expressed as spatially averaged PDPs, our use of local smoothing allows us to apply it to instantaneous PDPs with considerable success. Compared to previous work, our algorithm has several key features that contribute to its success: (1) We focus on the manner in which clusters of MPCs of given start time, duration, and exponential decay profile introduce discontinuities in the shape of the entire CIR. (2) We employ recursive partitioning to dramatically reduce the number of cluster combinations that must be checked and thereby make the algorithm tractable. (3) We enforce the assumptions of the S-V model through appropriate application of penalties. (4) We layer the cluster selection rules and take advantage of the iterative nature of the 58  algorithm so that it can be used as the basis for an interactive tool for use by analysts. Compared to current manual approaches, the proposed rule-based clustering algorithm helps to make cluster identification more consistent and less time-consuming. When used as the basis of an interactive graphical cluster identification tool, the algorithm should be a useful aid for those engaged in analysis of either conventional wideband or UWB channel impulse responses.  59  3.8  References  [1]  A. F. Molisch, “Ultrawideband propagation channels - Theory, measurement, and modeling,” IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1528-1545, Sep. 2005.  [2]  A. A. M. Saleh and R. A. Valenzuela, “A statistical model for indoor multipath propagation,” IEEE J. Sel. Areas Commun., vol. 5, no. 2, pp. 128-137, Feb. 1987.  [3]  A. F. Molisch, J. R. Foerster and M. Pendergrass, “Channel models for ultrawideband personal area networks,” IEEE Wireless Commun., vol. 10, no. 6, pp. 14-21, Dec. 2003.  [4]  A. F. Molisch et al., “IEEE 802.15.4a channel model – final report,” Tech. Rep., IEEE 802.15-04-0662-00-004a, Nov. 2004.  [5]  A. F. Molisch et al., “A comprehensive standardized model for ultrawideband propagation channels,” IEEE Trans. Antennas Propag., vol. 54, no. 11, pp. 31513166, Nov. 2006.  [6]  U. Schuster and H. Bolsckei, “Indoor UWB channel measurements from 2 GHz to 8 GHz,” Tech. Rep., IEEE 802.15-04-0447-00-004a, Aug. 2004.  [7]  J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, “A measurementbased statistical model for industrial ultra-wideband channels,” IEEE Trans. Wireless Comminic., vol. 6, no. 8, pp. 3028-3037, Aug. 2007.  [8]  D. Shutin and G. Kubin, “Cluster analysis of wireless channel impulse responses with hidden Markov models,” Proc. IEEE ICASSP, pp. 949-952, May 2004.  [9]  O.H. Woon and S. Krishnan, “Identification of clusters in UWB channel modeling,” in Proc. IEEE VTC 2006 - Spring, pp. 1-5, May 2006.  [10] M. Corrigan, A, Walton, W. Niu, and J. Li, “Automatic USB Cluster Identification,”, Proc. IEEE RWS ’09, pp. 376-379, Jan. 2009. [11] A. Massouri, J. Chen, L. Clavier, P. Combeau and Y. Pousset, “Automated Identification of Clusters and UWB Channel Parameters Dependancy on Tx-Rx Distance,” in Proc. IEEE EuCAP’09, pp. 3663-3667, Mar. 2009. [12] V. Erceg et al., “TGn Channel Models,” IEEE 802.11-03/940r4, May 2004. [13] L. J. Greenstein, S. S. Ghassemzadeh, S-C Hong and V. Tarokh, “Comparison study of UWB indoor channel models,” IEEE Trans. Wireless Commun., vol. 6, no. 1, pp. 128-135, Jan. 2007. [14] C-C. Chong and S. K. Yong, “A generic statistical-based UWB channel model for high-rise apartments,” IEEE Trans. Antennas Propag., vol. 53, no. 8, pp. 2389-2399, Aug. 2005. [15] J. Chuang, S. Bashir and D. G. Michelson, "Automated Identification of Clusters in UWB Channel Impulse Responses," in Proc. IEEE CCECE 2007, pp. 761-764, April 2007.  60  [16] A. F. Molisch, U. G. Schuster and C-C. Chong, “Measurement procedure and methods on channel parameter extraction,” Tech. Rep., IEEE 802.15-04-0283-00004a, May 2004. [17] I. Gijbels, A. Lambert and P. Qiu, “Edge-preserving image denoising and estimation of discontinuous surfaces,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 28, no. 7, pp. 1075-1087, Jul. 2006. [18] J. H. Friedman, “Multivariate adaptive regression splines,” The Annals of Statistics, vol. 19, no. 1, pp. 1-141, Mar. 1991. [19] M. G. diBenedetto, T. Kaiser, A. F. Molisch, I. Oppermann, C. Politano and D. Porcino, Eds., UWB Communications Systems: A Comprehensive Overview. Hindawi: EURASIP, 2005. [20] A. F. Molisch, "Ultra-Wide-Band Propagation Channels," Proc. IEEE, vol.97, no. 2, pp. 353-371, Feb. 2009. [21] S. Venkatesh, J. Ibrahim and R. M. Buehrer, “A new two-cluster model for indoor UWB channel measurements,” in Proc. IEEE AP-S Int. Symp. Dig., pp. 946-949, Jun. 2004.  61  Chapter 4 An Interim Propagation Model for Indoor Industrial Environments 4 4.1  Introduction  Encouraged by the success of wireless technologies in home and office environments, interest in wireless communication in industrial environments has steadily increased over the past two decades. The many benefits of wireless, rather than wired, communication presented in [1]–[3], however, the lack of a suitable model for wireless propagation in indoor industrial environments has hindered development and deployment. The performance limitations of any wireless system are determined by the environment it operates in, hence, propagation models that capture the prominent characteristics of the respective environment are required. These models are fundamental in enabling system designers to adequately engineer wireless systems. Although site-specific propagation models assist end-users in optimizing system deployment, site-independent propagation models are required by system designers to adequately plan and set performance expectations. Statistical site-independent models allow for generic simulation and testing of wireless systems and reveal vital trends without emulating the deterministic site-specific system behavior. In many cases, wireless devices in industrial environments are deployed as part of a large sensor network and the number of devices is of the order of the hundreds, or even thousands, which makes the computation of deterministic models extremely difficult. Over the last twenty years, only a handful of researchers have presented measurement results, and subsequently proposed statistical models, for wireless propagation in indoor industrial environments. As presented in [4]–[9], Rappaport and McGillem published a series of papers that were the first to detail wireless propagation in industrial environments. The papers were based on extensive UHF (1300 MHz) measurements collected in five factories in Indiana and included both narrowband and some wideband results. In the ISM A version of this chapter will be submitted for publication: A. Stefanski and D. G. Michelson, “An Interim Propagation Model for Indoor Industrial Environments.”  62  bands used for wireless LAN, [10] reported rms delay spread measurements at 5.2 GHz and [11] reported path-loss measurements at 2.45 GHz. More recently, [12] presented some results on large-scale and temporal fading comparing three of the ISM bands. With the emergence of UWB and its resilience to multipath interference, Karedal and Molisch published measurement results and a statistical model based on measurements campaigns in two industrial halls [13]–[18]. The preliminary results from those campaigns were used by the 802.15.4a channel modeling committee [16]–[17]. Clearly, the limited number of measurements campaigns means the results are still preliminary and make it difficult to draw general site-independent conclusions. In this chapter, we propose an interim propagation model based on existing literature to yield a set of channel model parameters for indoor industrial environments that span the license-exempt ISM and UWB bands ranging in frequencies from 800 MHz to 10.6 GHz. Our model includes parameters for the distance and frequency dependence of pathloss, the statistical distribution of shadowing, the modeling of channel impulse response and delay spread, and also small-scale fading. In statistics, this process is referred to as meta-analysis and such rationalized modeling techniques have been used previously, for example, in [24], for modeling the distribution of rms delay spread in cellular environments and, more recently, in [25] by the 802.15.3c channel modeling sub-committee to aid the development of an millimeter-wave based alternate physical layer. Furthermore, parts of the industrial environment results described in 802.15.4a report were extracted in a similar fashion. The remainder of the chapter is organized as follows: in Section 4.2 we provide a comprehensive summary of the existing propagation measurement campaigns in indoor industrial environments. In Section 4.3, we present a summary of the typical propagation scenarios in industrial environments and also provide the form of the interim propagation model based on the existing knowledge about path-loss and shadowing, the shape of the channel impulse response and delay spread, and also small-scale fading. Next, in Section 4.4 we present the empirical parameters for our rationalized channel model. Finally, in Section 4.5, we present a summary and draw conclusions.  4.2  Summary of Existing Measurements  Previous UHF, ISM, and UWB indoor industrial propagation measurement campaigns are summarized in Table 8. The first propagation measurements in industrial environments 63  were reported by Rappaport in [4] and were subsequently used for [5]–[9]. The measurements were done using a signal generator (and a pulse generator the wideband measurements) and digital oscilloscope at UHF frequencies of approximately 1300 MHz. The frequency selection was a compromise between the 800/900 MHz and 1500/1700 MHz bands. The propagation measurements were taken at five fully operational factories that varied in size and layout and the separation distance was varied between 10m and 80m. The measurement data allowed for the extraction of various narrowband channel parameters and also for power delay profiles for the characterization of multipath channel parameters. In all, the campaign resulted in over 20,000 measurements at over 50 locations. In the ISM band, three measurements campaigns have been reported in the open literature. The first was reported in [10] and were carried out measurements at 5.2 GHz in a car manufacturing hall and an aircraft assembly hanger. The measurements were done using a RUSK channel sounder at distances between 0 and 140m and delay and angular spread parameters were extracted. [11] presented received power results using IEEE 802.11 compliant equipment in the ISM bands. The measurements were taken at 2.45 GHz using wireless local area network equipment at three different industrial environments: a chemical pulp factory, a cable factory, and a nuclear power plant. More recently, [12] presented results comparing propagation at the three ISM bands: 900 MHz, 2400 MHz, and 5200 MHz. The literature presents path-loss and temporal fading results from channel measurements taken in two wood processing plants and two metal factories. Lastly, in UWB there have been four measurements campaigns have been reported. In [13]–[14], Karedal and Molisch published the first UWB indoor industrial measurements based on 10 measurement locations in a small incinerator hall in Sweden. A measurementbased statistical model for industrial UWB channel was presented in [18], however, the authors acknowledge the dataset to be limited and do not claim a ‘general’ model. [19] developed a simple and flexible locally-coherent UWB model for industrial environments that generalized the geometry-based stochastic channel model, a concept that is widely used in wideband channel models. In [20], measurement results in a laboratory representative of a process industry environment and parameters for pathloss and time dispersion were provide. Finally, [21] presented results for 0.8 to 4.0 GHz measurements at 10 locations within a wood processing facility. As one can see, the UWB measurement campaigns all have been relatively small and the results limited. 64  TABLE 8: SUMMARY OF INDOOR INDUSTRIAL PROPAGATION MEASUREMENT CAMPAIGNS Author [Yr]  Ref.  Environ.  Frequency  Method  Path Range  No. of Sites [Loc.]  Parameters Extracted Path Loss, RMS Delay Spread, Mean excess delay, Large scale fading, number of MPC, small scale fading, spatial and temporal correlation functions  1.3 GHz  Pulse generator w/ DSO  10 to 80 m  5 sites w/ 50 total measurement locations (~6 sites, 4 scenarios, 3 locations)  Rappaport  [4]–[8]  Office Building, Food Processing Plant, Engine Manufacturing Plant, Aluminum Manufacturing Facility, Casting Foundry, Engine Machine and Assembly Shop  Yagani  [9]  same dataset above  as above  as above  as above  as above  Hampicke  [10]  Car manufacturing hall (100m x 20m) and aircraft hanger (300m x 120m)  5.2 GHz  RUSK Sounder  0 to 140m  2 sites  Kjesbu  [11]  Chemical Pulp Factory, Cable Factory, and Nuclear Power Plant  2.45 GHz  CW Wave w/ Spectrum Analyzer  0 to 95 m  3 sites w/ 2 paths per site  RSSI, slow fading, RMS delay spread, Ricean KFactor  Tanghe  [12]  4 typical factories (two wood factories and two metal processing factories) (narrow-band)  900, 2.4 GHz, 5.2 GHz  Sig. Gen w/ Spectrum Analyzer  15 m to 140 m  960 collected samples (4 factories, 4 topos, various distances)  Path Loss, standard deviation of PL, temporal fading, Rician K-Factor  Karedal  [13] [14] [18]  small incinerator hall (small - 13.6 x 9.1 x 8.2) LOS, NLOS, and BS-NLOS  3.1 to 10.6  VNA  2, 4, 8m  1 site, 10 locations  Molisch  [15] [16]  From Karedal’s Measurements  3.1 to 10.6  VNA  2, 4, 8m  1 site, 10 locations  Khan  [17]  MAX-Lab in Sweden (medium size 94x 70m- Electron Accelerator Laboratory)  3.1 to 8 GHz  VNA  Kunisch & Pamp  [19]  Hall w/ some machinery (32m x 24.5m x 8.5m) (A-A = LOS and S-A = NLOS)  3 - 11 GHz  VNA  Irahhauten  [20]  representative for an actual "Process Industry" radio environment LOS and NLOS  12 GHz  Pulse Gen  Tanghe  [21]  Wood processing facility (7m height)  0.8 – 4 GHz  VNA  UHF  ISM  UWB  2-8m and 10-16m < 20m (LOS) & <10m (NLOS)  10 + 6 locations 1 site  CIR statistical model, small scale fading Delay spread, azimuth spread  Path Loss, RMS delay spread, S-V Parameters, small-scale fading Path Loss, RMS delay spread, S-V Parameters, small-scale fading Path Loss, RMS delay spread, NP of MPCs Path Loss and scattering statistics  1-25m  73 CIRs  Path Loss, standard deviation of PL, RMS delay spread  10 to 35m  5 LOS and 5 NLOS  S-V model parameters  65  4.3  Rationalized Channel Parameters  4.3.1  Classification of Propagation Topographies  Channel model parameters will vary depending on the propagation environment. Based upon [5], our rationalized channel model classifies environments based upon the presence, or obstruction, of a direct line-of-sight (LOS) path and upon the amount of metallic clutter present in the surrounding environment. For the purpose of this chapter, we refer to all links without a LOS component present as obstructed, even if the cited literature refers to the link as non-line-of-sight (NLOS). Line-of-Sight with Light Clutter – This scenario characterized by the presence of an LOS path and the antennas are surrounded with little to no metallic clutter. The clutter is typically located at, or below, the height lower than then the antennas. Examples of this scenario are open areas, low density work areas (ie: machine shops), and major aisle by relatively empty storage areas. Line-of-Sight with Heavy Clutter – This scenario characterized by the presence of an LOS path and the antennas are surrounded with heavy metallic clutter. The clutter is typically near, or above, the respective antennas. Examples of this scenario are automated assembly lines and large manufacturing and/or process factories. Obstructed with Light Clutter – This scenario characterized by the obstruction of the LOS path and the antennas are surrounded with low density clutter. This occurs in a light clutter environment as described above, however, the dominant direct component is obstructed and only the reflections are captured by the receive antenna. Examples of this scenario are machine shops and assembly lines. Obstructed with Heavy Clutter – This scenario characterized by the obstruction of an LOS path and the antennas are surrounded with heavy metallic clutter. An example of this scenario is propagation across factory and even between adjacent aisles. A fifth scenario that has sometimes been reported is literature is that of a single sided aisle which can be typically found around the perimeter of an industrial factory. As reported in [7], this topography has propagation characteristics similar to that of hallways, hence, we do not consider it as a unique scenario. Also, multi-storied buildings are not treated as a special case as most factories are either single storied or have thick floors 66  concrete floors, required to support heavy machinery, which severely attenuate the wireless signal. 4.3.2  Path-Loss and Shadowing  The path-loss for our rationalized channel model takes the form of the narrowband path loss model. It is typically used for statistical analysis of the log-distance and can be expressed in decibels (dB) as  d PL dB ( d ) = PL 0 + 10 n log 10   d0    + 20 log f + X σ ,   (4.1)  where PL0 is the mean path loss at a reference distance, n is the topology-specific path-loss exponent, f is the frequency, and Xσ is a random variable for modeling the shadowing (i.e., large-scale fading). The random variable for shadowing is assumed to take the form of a zero-mean Gaussian distribution with the standard deviation, σ. This results in a log-normal distribution since Eqn. (1) is expressed in dB. For UWB communication, it is also necessary to consider the frequency dependence of the pathloss as the attenuation at low frequencies can be significantly different than that at high frequencies. For simplicity, the two are assumed to be independent of each other and are the product of the individual terms, PL dB ( d , f ) = PL dB ( d ) + PL dB ( f ) .  (4.2)  Following [15], the frequency dependence of path loss for wideband system can be expressed using the simple power law,  PL( f ) ∝ f −2κ ,  (4.3)  where κ is the monotonically decaying slope of the frequency channel response and can be determined by least-square fitting methods. It is also worth noting that the Equ. (3) above is independent of antenna frequency effects that need to be considered for UWB systems. 4.3.3  Channel Impulse Response  4.3.3.1 Power Delay Profile Assuming that the propagation channel is time-invariant, the impulse response channel response for the rationalized channel model is a tapped delay line given by  67  N −1  h(τ ) = ∑akδ (τ −τ k )e jθk ,  (4.4)  k =0  where δ(τ) is the dirac function, τk is the arrive time of the multipath component, ak is the path amplitude, θk is the path phase, and N is the number of resolvable multipath components. Typically, the phase is assumed to be uniformly distributed between [0, 2π) and is omitted. The power delay profile (PDP) is the square magnitude of the impulse response, N −1  PDP(τ ) = h(τ ) = ∑ ak δ (τ − τ k ) . 2  2  (4.5)  k =0  4.3.3.2 Delay Spread Although the equations above provide a complete description of the channel impulse response, it is also beneficial to include descriptions for RMS delay spread (τrms), the mean excess delay (τmean), and the number of dominant paths. These parameters help provide useful information to system designers about the shape, duration, and structure of the CIR. For example, the ratio of the RMS delay spread to the symbol period is usually strongly correlated to the intersymbol interference (ISI) and biterror-rate (BER). The number of dominant paths is an important consideration for RAKE receiver design and provides insight into how many matched filters are required for capturing a certain amount of energy. The excess delay is a measure of the delay between the kth multipath component and the first arriving path. Hence, the normalized first-order moment of the power delay profile yields the mean excess delay and is expressed as ∞  ∫ ∫  −∞ ∞  τ mean =  −∞  PDP(τ )τdτ  (4.6)  PDP(τ )dτ  and the RMS delay spread is defined as the second central moment of the power delay profile, τ rms =  ∫  ∞  PDP (τ )τ 2 d τ  −∞ ∞  ∫  −∞  PDP (τ )dτ  2 − (τ mean ) .  (4.7)  For simplicity, the distance dependence of the delay spread can be ignored, however, following [20], it can be modeled by the power law with a constant slope, 68  τ rms ∝ d c .  (4.8)  The number of dominant paths is simply the number of multipath components above a certain threshold (ie: NP20dB). A variant of this is the number of dominant paths required to captures a certain amount of energy (ie: NP85%). 4.3.4  Small-Scale Fading  Small-scale, or fast, fading describes the amount the amplitude of a signal can change with relatively small, in the order of λ/2, changes in spatial position. The probability distributions commonly used to model small-scale fading include Rayleigh, lognormal, Nakagami, Rice, and Wiebull. To model the small-scale fading in industrial environments, we selected to use the Nakagami distribution as it affords a large amount of flexibility. The Nakagami distribution is given by the following equation, m  2  m  2 m−1  m  f x (x ) = exp − x 2  ,   x Γ(m )  Ω   Ω   (4.9)  where m ≥ ½ is the Nakagami m-factor (i.e. the shape parameter of the distribution), Γ(m) is the Gamma function, and Ω is the mean-squared value of the amplitude (i.e. the spread parameter of the distribution). The advantage of the Nakagami distribution is that it can be generalized to other common distributions. For example, for m = 1 the distribution becomes Rayleigh and for m >> 1 the distribution becomes Lognormal. Furthermore, as described in [15], it is also possible to convert from a Nakagami distribution to an approximation of Rice distribution with the following conversion equations: 2 ( Kr + 1) , and m=  (4.10)  2K r + 1  Kr =  m2 − m  ,  m − m2 − m  (4.11)  where Kr and m and the Rice and Nakagami-m factors, respectively. When not generalized, the m-parameter is modeled as a lognormal distributed random variable whose logarithm has mean m0 and standard deviation  m̂0 ,  however, the LOS component is modeled as  ~ . m 0  69  4.4  Results  In the following sections, we provide empirical values and supporting discussions based upon the measurement literature summarized in Table 8. For the summary tables below (i.e. Table 9–11), we have not added empirical results have not been reported in literature, however, the reader may approximate values based upon values provide. 4.4.1  Path-Loss and Shadowing  The large number of available path-loss measurements, relative to other channel parameters in industrial environments, allowed for meaningful extraction of most of the path-loss parameters. For line-of-sight scenarios, the path-loss exponent was n = 1.7 and n = 1.4, for light and heavy clutter, respectively, as shown in Figure 19. For obstructed scenarios, the path-loss exponent was n = 1.8 and n = 2.9, for light and heavy clutter, respectively, as shown in Figure 20. The path-loss exponent being lower than 2 (i.e., the exponent for free-space) is due to the multipath propagation that is present in industrial environments. When reporting PL0, it is possible to use fixed (i.e., relative to free-space path-loss) and non-fixed intercept models. As detailed in [12], non-fixed intercept provide a significantly better fit to measured data. From empirical data, we found that the non-fixed intercept is typically 7 to 13dB lower than then fixed intercept. For frequency dependence, we found that the exponent was varied from κ = 0.91 to κ = 1.30, depending on the scenario. Lastly, the standard deviation was typically reported to vary between 3 to 8 dB. A summary of the path-loss parameters is provided in Table 9. The PL0 reference distance 1m. As mentioned above, the values have been derived from measured data ranging from 0m to more than 100m and at frequencies from 800 MHz to 10.6 GHz. TABLE 9: SUMMARY OF PATH-LOSS PARAMETERS  PL0 n σ κ  Line-of-Sight Light Heavy 62.6 dB 1.7 1.4 3.8 dB 4.4 dB 1.30  -  Obstructed Light Heavy 73.1 dB 75.3 dB 1.8 2.9 3.7 dB 3.9 dB 1.09  0.91  70  Plot of Line-of-Sight Path Loss vs. Distance 0  -5  -10  Path Gain [dB]  -15  -20  -25  -30  -35  -40 0 10  Rappaport 1989 - LOS Kjesbu 2000 - Light LOS Kjesbu 2000 - Heavy LOS Molisch 2006 - Light LOS Irahhauten 2008 - Light LOS Tanghe 2008 - LOS Best Fit - Light LOS Best Fit - Heavy LOS 1  2  10  10  Distance [m]  Figure 19: Plot of Line-of-Sight Path Loss vs. Distance Plot of Obstructed Path Loss vs. Distance 0 Rappaport 1989 - Light OBS Rappaport 1989 - Heavy OBS Khan 2005 - Heavy OBS Khan 2005 - Light OBS Irahhauten 2008 - Heavy OBS Tanghe 2008 - Light OBS Tanghe 2008 - Heavy OBS Best Fit - Light OBS Best Fit - Heavy OBS  -5  -10  Path Gain [dB]  -15  -20  -25  -30  -35  -40 0 10  1  10  2  10  Distance [m]  Figure 20: Plot of Obstructed Path Loss vs. Distance  71  4.4.2  Channel Impulse Response  Due to the limited number of measurements available, only a summary of the wideband and UWB channel impulse response parameters is provided in Tables 10 and 11, respectively. The wideband parameters can be used between 900 MHz and 5.2 GHz, whereas, the UWB parameters can be used for frequencies ranging between 800 MHz and 10 GHz. For the UWB case, modified S-V parameters are also provided since the model was selected by the 802.15.4a channel modeling committee. TABLE 10: SUMMARY OF WIDEBAND CHANNEL IMPULSE RESPONSE PARAMETERS  τrms NP20dB  Line-of-Sight Light Heavy 75 ns 78 ns 14  -  Obstructed Light Heavy 91 ns 83 ns 30  -  TABLE 11: SUMMARY OF UWB CHANNEL IMPULSE RESPONSE PARAMETERS  1/Λ Γ γ γ (τ) τrms NP20dB NP85%  4.4.3  Line-of-Sight Light Heavy 15.8 ns 12.6 ns 3.5 ns 0.8 32 ns ~50 ~40 -  Obstructed Light Heavy 14.3 ns 27.5 ns 3.9 ns 0.8 ns 90 ns > 1250 > 1000  Small-Scale Fading  Small-scale fading affects the mobile wireless channel due to rapid amplitude changes cause by motion. This information is critical for designing asset tracking or localization systems. At lower bandwidths, [5] found that the receive signal can experience very deep fades, up to 30 dB, and it not correlated with scenario or path distance. For LOS scenarios, where a large portion of the receive power dominates the envelope of the power delay profile, the distribution is typically Rician. For obstructed scenarios, the fading distribution is typically lognormal. However, for UWB signals, as reported in [18] and [21], the power delay profiles are Rayleigh distributed (ie: m = 1) due to the large number of scatters present in industrial environments, expect for the LOS path which is Nagakami distributed. 72  4.5  Conclusions  The lack of a comprehensive propagation model for in indoor industrial environments has hindered development and deployment. As we have shown, the existing indoor industrial propagation models have been mostly produced based on anecdotal evidence. Without a clear framework in place, previous researchers have for the most part presented sitespecific models that make it difficult to draw broad conclusions about the environment and a complete representation of the propagation channel will only emerge when a large body of measurement evidence spreading ISM and UWB frequencies is available We have proposed a site-independent rationalized channel model based on existing literature that is suitable for use in indoor industrial environments between 800 MHz and 10.6 GHz. Our model takes the form of statistical equations that describe the path-loss and shadowing, shape of the channel impulse response, and also small-scale fading. We have also identified four typical propagation scenarios as wireless propagation is highly dependent on the operating environment. Lastly, based on existing measurements, we have included parameters for our proposed model. Our results help identify the research areas of interest in industrial environments and future measurement campaigns will optimize and tune the model. Ultimately, by increasing the understanding of wireless propagation in industrial environments, the model will assist system designers in adequately determining performance expectations and creating guidelines that enable users to realize more efficient system planning, site assessment, and improved deployment of access points  73  4.6  References  [1]  A.Willig, K. Matheus and A.Wolisz, “Wireless technologies in industrial networks,” Proc. IEEE, vol. 93, no. 6, pp. 1130–1150, Jun. 2005.  [2]  A. Willig, "Recent and Emerging Topics in Wireless Industrial Communications: A Selection," IEEE Trans. Ind. Informat., vol. 4, no. 2, pp. 102-124, May 2008.  [3]  G. P. Hancke and B. Allen, "Ultrawideband as an Industrial Wireless Solution," IEEE Pervasive Comput., vol. 5, no. 4, pp. 78-85, Dec. 2006.  [4]  T. S. Rappaport and C. D McGillem, “UHF multipath and propagation measurements in manufacturing environments,” in Proc. IEEE GLOBECOM, pp. 825-831, 1988.  [5]  T. S. Rappaport and C. D McGillem, “UHF fading in factories,” IEEE J. Sel. Areas Commun., vol. 7, no. 1, pp. 40-48, Jan. 1989.  [6]  T. S. Rappaport, "Indoor radio communications for factories of the future," IEEE Commun. Mag., vol. 27, no. 5, pp. 15-24, May 1989.  [7]  T. S. Rappaport, "Characterization of UHF multipath radio channels in factory buildings," IEEE Trans. Antennas Propag., vol. 37, no. 8, pp.1058-1069, Aug. 1989.  [8]  T. S. Rappaport, S. Y. Seidel and K. Takamizawa, “Statistical channel impulse response models for factory an open plan building radio communication system design,” IEEE Trans. Commun., vol. 39, no. 5, pp. 794–807, May 1991.  [9]  P. Yegani and C.D. McGillem, "A statistical model for the factory radio channel," IEEE Trans. Commun., vol. 39, no. 10, pp. 1445-1454, Oct. 1991.  [10] D. Hampicke, A. Richter, A. Schneider, G. Sommerkom, R.S. Thoma and U. Trautwein, "Characterization of the directional mobile radio channel in industrial scenarios, based on wideband propagation measurements," in Proc. IEEE VTC’99, 1999, pp.2258-2262. [11] S. Kjesbu and T. Brunsvik, "Radiowave propagation in industrial environments," in Proc. IEEE IECON, pp. 2425-2430, Oct. 2000 [12] E. Tanghe, W. Joseph, L. Verloock, L. Martens, H. Capoen, K. Van Herwegen and W. Vantomme, "The industrial indoor channel: large-scale and temporal fading at 900, 2400, and 5200 MHz," IEEE Trans. Wireless Commun., vol. 7, no. 7, pp. 27402751, Jul. 2008. [13] J. Karedal, S. Wyne, P. Almers, F. Tufvession and A. F. Molisch, "Statistical analysis of the UWB channel in an industrial environment," in Proc. IEEE VTC’04, pp. 81-85, Sep. 2004, [14] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "UWB channel measurements in an industrial environment," in Proc. IEEE GLOBECOM, pp. 35113516, Dec. 2004. [15] A. F. Molisch et al., “IEEE 802.15.4a Channel Model Final Report,” Tech. Rep., IEEE 802.1504-0062-02-004a, 2005.  74  [16] A. F. Molisch et al., "A Comprehensive Standardized Model for Ultrawideband Propagation Channels," IEEE Trans. Antennas Propag., vol. 54, no. 11, pp. 31513166, Nov. 2006. [17] M. G. Khan, A. A. Ashraf, J. Karedal, F. Tufvesson, and A. F. Molisch, “Measurements and Analysis of UWB Channels in Industrial Environments,” in Proc. WPMC, Sep. 2005. [18] J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "A MeasurementBased Statistical Model for Industrial Ultra-Wideband Channels," IEEE Trans. Wireless Commun., vol. 6, no. 8, pp. 3028-3037, Aug. 2007. [19] J. Kunisch and J. Pamp, "Locally Coherent Ultra-Wideband Radio Channel Model for Sensor Networks in Industrial Environment," in Proc. ICUWB, pp. 363-368, Sep. 2006. [20] Z. Irahhauten et al., "UWB Channel Measurements and Results for Office and Industrial Environments," in Proc. IEEE ICUWB’06 , pp. 225-230, Sep. 2006. [21] E. Tanghe, W. Joseph, J. De Bruyne, L. Verloock and L. Martens, "The Industrial Indoor Channel: Statistical Analysis of the Power Delay Profile,” in Int. J. of Elect. Commun., vol. 64, no. 9, pp. 806-812, Sep. 2010. [22] L. J. Greenstein, V. Erceg, Y.S. Yeh and M.V. Clark, "A new path-gain/delay-spread propagation model for digital cellular channels," IEEE Trans. Veh. Technol., vol. 46, no. 2, pp. 477-485, May 1997 [23] S-K. Yong et al., “TG3c Channel Modeling Sub-committee Final Report,” Tech. Rep., IEEE802.1507-0584-01-003c, 2007.  75  Chapter 5 Conclusions and Recommendations The work presented in this thesis represents three contributions to the field of wireless propagation in indoor industrial environments. Here, we summarize these contributions, assess their limitations and offer recommendations for future work. To the best of our knowledge, we have presented the first comprehensive review of both existing and potential applications for UWB wireless technology in indoor industrial environments. Short-range wireless technology is a proven method for: (1) simplifying connection to mobile and rotating assets, (2) reducing cabling and re-configuration costs, (3) increasing reliability by introducing redundant paths and (4) substantially reducing the risk of cable or connector damage [1],[2]. Compared to conventional short-range wireless technologies, UWB wireless technology offers: (1) greater resistance to multipath fading and interference, (2) potentially lower power consumption, (3) support for higher data rates and (4) provision for accurate measurement of distance, velocity and position, and intrusion [3]. Thus, we conclude that UWB wireless devices and networks have the potential to play a significant role in industrial applications in the form of: (1) cable replacements, (2) physical sensors and (3) sensor networks. The lack of propagation models suitable for use in system design or evaluation increases risk for developers and is a significant impediment to progress. While our survey is necessarily limited to information that we are able to gather from the open academic and trade literature, the results will help future researchers more easily put their own contributions to this field in context. We have demonstrated that efforts to characterize UWB channel impulse responses (CIRs) in new environments will benefit from a computer-assisted cluster identification algorithm for fitting measured CIRs to the Saleh-Valenzuela (S-V) model that is potentially more consistent and less time-consuming than current manual approaches. We express the corresponding power delay profile (PDP) on a semi-logarithmic scale so that exponential decay profiles that correspond to individual clusters will appear as straight lines. We use linear regression to determine the particular combination of straight lines that best fit the PDP in a least squares sense. The lines correspond to clusters while their slopes give the 76  intracluster decay rate. If the fit does not satisfy our RMS error criterion, we increment the number of clusters by one and repeat the process. Assuming that adding a cluster will always involve subdividing an existing cluster dramatically reduces the number of combinations that must be checked and makes the algorithm tractable. The analyst interactively enforces the assumptions of the S-V model through appropriate application of penalties. Minor extensions to the ideal S-V model can be accommodated through appropriate pre- and post-classification tuning by the analyst. Cluster identification trials conducted using both simulated and measured UWB CIRs confirm the robustness and practicality of our approach. Designers of short-range wireless networks intended for deployment in indoor industrial environments require propagation and channel models that can be used to fairly compare alternative system configurations under realistic conditions [4]. Although only a handful of measurement campaigns have been reported in a variety of bands between 800 MHz and 10.6 GHz, no comprehesive propagation model exists for indoor industrial environments. Here, we have carefully considered the results obtained in each previous campaigns in order to generate a rationaized model that captures path loss and shadowing, fading and frequency dispersion, and time dispersion across the range 800 MHz to 10.6 GHz in selected deployment scenarios. The four deployment scenarios include both line-of-sight and non-line-of-sight paths in both heavy and light clutter. We also identify gaps in previous work that need to be filled. The results will be useful to both practitioners who require access to site-general models for analysis and design and researchers who wish to contribute to the further development of such models.  77  5.1  References  [1]  A.Willig, K. Matheus and A.Wolisz, “Wireless technologies in industrial networks,” Proc. IEEE, vol. 93, no. 6, pp. 1130–1150, Jun. 2005.  [2]  A. Willig, "Recent and Emerging Topics in Wireless Industrial Communications: A Selection," IEEE Trans. Ind. Informat., vol. 4, no. 2, pp. 102-124, May 2008.  [3]  G. P. Hancke and B. Allen, "Ultrawideband as an Industrial Wireless Solution," IEEE Pervasive Comput., vol. 5, no. 4, pp. 78-85, Dec. 2006.  [4]  A. F. Molisch, "Ultra-Wide-Band Propagation Channels," Proc. IEEE, vol. 97, no. 2, pp.353-371, Feb. 2009.  78  Appendices Appendix A: Cluster Identification MATLAB GUI This appendix provides a detailed description of the computer-assisted cluster identification software that we used to generate the results presented in Section 3.3. The software was written in MATLAB and implements a graphical user interface (GUI) that allows the analyst to interact with the automated cluster identification algorithm.  Step 1: Initial Processing The first step implemented by the software is initial processing. Here, the analyst can set and adjust the amplitude and delay cutoffs to eliminate noise or any unwanted portions of the power delay profile. The analyst has options to set the amplitude threshold (dB) and the maximum delay (ns). Although only the Saleh-Valenzuala model is used for fitting UWB data in Chapter 3, two other common models have been implemented in the software, i.e., exponential and exponential-lognormal. A screenshot of the initial processing step is shown in Figure 21.  Figure 21: Screenshot of the Initial Processing Step  79  Step 2: Local Smoothing The second step of the algorithm is local smoothing, as described in Section 3.3.1. In this step, the analyst can smooth the channel impulse response by setting the smooth interval, N, and by setting the decimate factor, k. Furthermore, the analyst is able to see an initial estimate for the RMS error for linear exponential decay of the cluster. Lastly, one can remove any outliers and select the MPC definition to include all taps or local maximums. A screenshot of the local smoothing step is shown in Figure 22.  Figure 22: Screenshot of the Local Smoothing Step  Step 3: Cluster Identification The third step of the algorithm is cluster identification, as described in Section 3.3.1. Using the methods described in Sections 3.3.2 to 3.3.4, the cluster identification algorithm proceeds to identify the start time of clusters. Due of the difficulty of developing a completely autonomous algorithm, the analyst must adjust the default penalty values to 80  enforce the shape of the S-V model as defined by the 802.15.4a channel modeling subcommittee. A screenshot of the cluster identification step is shown in Figure 23.  Figure 23: Screenshot of the Cluster Identification Step  Step 4: Final Adjustments The fourth step of the cluster identification algorithm GUI is final adjustments. Here, the analyst is able to extract the model data and save the desired model parameters to a text output file for further analysis. Furthermore, two additional features are implemented. The analyst is able to fit a power law curve to individual cluster, rather than an exponential decay, described in [1]. Also, the analyst is able to manual adjust the start time of the individual clusters is the algorithm was not able to identify them successfully. A screenshot of the final adjustments step is shown in Figure 24.  81  Figure 24: Screenshot of the Final Adjustments Step  82  References [1]  J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, "A MeasurementBased Statistical Model for Industrial Ultra-Wideband Channels," IEEE Trans. Wireless Commun., vol. 6, no. 8, pp. 3028-3037, Aug. 2007.  83  Appendix B: UWB Channel Sounder This appendix provides details of the measurement campaign described in Section 3.5.2. It provides a more detailed explanation of both the UWB Channel Sounder and the UWB measurement campaigns it enabled.  UWB Channel Sounder The objective of UWB channel measurement campaigns is to determine the channel impulse response within propagation environments of interest. UWB channel impulse responses can be determined using various measurement devices, such as impulse sounders [1], vector network analyzers [2],[3], and correlative channel sounders [4]. We performed frequency-domain channel sounding measurements of the indoor industrial environment UWB channel using the setup configuration shown in Figure 25.  Figure 25: Block Diagram of the UWB Channel Sounder  Our UWB channel sounder consists of an Agilent E8362B vector network analyzer (VNA) equipped with two Electro-Metric 6865 UWB omni-directional biconical antennas, a 0.5m-by-1.0m two-dimensional linear antenna positioner based on two Velmex BiSlide assemblies and VMX programmable step-motor controllers, a Miteq SCMT-100M11G optical transmitter with a matching Miteq SCMR-100M11G optical receiver, a MiniCircuits ZVA-183+ wideband amplifier, optical fiber and coaxial RF cables, and a laptopbased instrument controller equipped with RS-232 and GPIB interfaces. A MATLAB script was used for logging the measurement data and also for control of both the VNA and the antenna positioner during the measurement campaigns. The VNA was programmed to sweep over the entire FCC spectrum mask for UWB 84  transmission from 3.1 GHz to 10.6 GHz in 6401 linearly distributed frequency steps with an intermediate frequency bandwidth of 3 kHz. The 7.5 GHz bandwidth results in a temporal resolution of approximately 0.133 ns or, correspondingly, a spatial resolution of 0.040 m. The 1.172 MHz frequency resolution results in a maximum excess delay of 853.3 ns or, correspondingly, a maximum distance range of 255.8 m.  In order to avoid  measurement errors, the VNA was set to dwell for 0.5 ms at each frequency point prior to taking the response data rather than performing a typical analog frequency sweep. The latter method is not recommended for frequency-domain measurements since it is possible for an appreciable frequency shift to occur between the source and multipath components arriving at the receiver unless an appropriately slow sweep time is used. Throughout the measurement campaign, at each measurement location the transmit antenna is kept in a fixed position and the receive antenna is moved around a square 7x7 virtual array using the two-dimensional antenna positioner. A 7x7 array was selected as approximately 50 samples are required to have a relevant statistical sample. The x-axis and y-axis spacing between each virtual element was selected to be λ/2 of the lowest measured frequency (48.35 mm at 3.1 GHz) so that the fading is independent at each virtual element. Both the transmit and receive antennas were mounted at a height of 1.25 m above the ground level.  Data Pre-Processing After the raw measured data is exported from the VNA, it is necessary to apply a calibration process to compensate for the non-linear effects of the amplifiers and cables that connect the VNA to the transmitting and receiving antennas. This is accomplished by applying a through-line calibration and also removing the reflection losses for both the transmit and receive antennas. The resulting frequency response is often referred to the radio channel since the measurements include the propagation channel along with the effects of the antennas. The challenging task of de-embedding the propagation channel from the radio channel in UWB measurement has been documented in literature since all realizable antennas have a radiation pattern that does not weigh all directions and all frequencies equally. To completely de-embed the propagation channel response from the radio channel it is necessary to first know the radiation pattern of both of the antennas, which can be 85  accomplished via a calibration measurement in an anechoic chamber. For our UWB antennas, University of Manitoba provides use with such measurement. Next, it is necessary to measure the frequency-dependant double-directional channel response to determine the angle-of-departure (AoD) and the angle-of-arrival (AoA) using advanced algorithms such as CLEAN, ESPRIT, MUSIC, or SAGE algorithm.  However, these  algorithms are very complex and suffer from possible noise enhancement. Typically, the de-embedding procedure is simplified by either assuming an antenna transfer function that is averaged in all directions or only in the azimuthal plane. For the purpose of this chapter, we average in all directions since the metal machinery in industrial environments results in high amounts of scattering.  Measurement Sites UWB channel sounding measurements were conducted at various locations around the University of British Columbia (UBC) campus in Vancouver, Canada. The first site is an indoor office environment where the UBC Radio Science Laboratory is located and the UWB channel sounder was developed. The remaining sites are a diverse set of indoor industrial environments located around campus. Site #1 – MacLeod Building – The Electrical and Computer Engineering building is a 4 floor concrete building constructed in 1963. The majority of measurements were taken in classrooms (6m x 10m x 3m), hallways (3m x 38m x 2.5m), and lecture halls (11m x 18m x 3.5m) which have suspended ceilings. The remaining measurements were taken in a an atrium (12m x 25m) where half of them had a 2.5m wood paneled ceiling and the other half had a 16m glass ceiling. Site #2 – AMPEL High Headroom Laboratory – The Advanced Materials and Process Engineering Laboratory (AMPEL) High Headroom Laboratory provides space for large scale experiments such as casting of metals and alloys, preparation of ceramics, plasma spray oxide coatings, fluidization, and a steel run out mill. The floor area measures 1500 m2, half of which has a ceiling height of 8m with metal trusses and the other half has a 3m concrete ceiling. The laboratory includes a 5 tonne gantry crane, a 5.5m long rotary kiln, fluidized bed reactors, 20m long runout table, and other metallurgical processing facilities.  86  Measurement Database Our measurement database includes both development runs and production runs. For the development runs we collected data in an indoor industrial environment (29 LOS & 6 NLOS locations) and also an indoor office environment to perform consistency check and validate our measurement procedure. For our production runs, in the indoor industrial environment we took measurements at 45 different locations at distances varying between 1m and 18m. For the indoor office environment we took measurements at 40 different locations at distances varying between 1m and 18m. 49 spatial samples were taken at each measurement location, hence, these two sets combined to yield almost 6000 CIRs.  87  References [1]  A. F. Molisch, “Ultrawideband Propagation Channels-Theory, Measurement, and Modeling,” IEEE Trans. Veh. Technol., vol. 54, no. 5, pp. 1528-1545, Sep. 2005.  [2]  J. Karedal, S. Wyne, P. Almers, F. Tufvesson and A. F. Molisch, “A measurementbased statistical model for industrial ultra-wideband channels,” IEEE Trans. Wireless Communic., vol. 6, no. 8, pp. 3028-3037, Aug. 2007.  [3]  C-C. Chong and S. K. Yong, “A generic statistical-based UWB channel model for high-rise apartments,” IEEE Trans. Antennas Propag., vol. 53, no. 8, pp. 2389-2399, Aug. 2005.  [4]  W. Ciccognani, A. Durantini, ans D. Cassioli, "Time domain propagation measurements of the UWB indoor channel using PN-sequence in the FCC-compliant band 3.6-6 GHz," IEEE Trans. Antennas Propag., vol. 53, no. 4, pp. 1542- 1549, April 2005  88  Appendix C: UWB Channel Sounder MATLAB Code This appendix provides details of the measurement campaign described in Section 3.5.2. It provides original MATLAB source code used for controlling the VNA and for processing the frequency responses. The section is organized as follows: channelSounder.m  –  The main file that is executed when running the UWB channel Sounder  inputParameters.m  –  The header file that stores all of the input parameters and constants  positionerInit.m  –  The initialization file for configuring the 2D linear positioner  PNAInit.m  –  The initialization file for configuring the vector network analyzer  startMeasurement.m  –  Function for controlling the Rx Antenna position around virtual array and taking measurements at each position as required.  processRawData.m  –  Function for applying through-line calibration and antenna transfer  cfr2fir.m  –  Function for transforming channel frequency response to channel impulse response  channelSounder.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % UWB Channel Sounder % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - June 12, 2008 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Clean up workspace clear all; close all; clc; delete(instrfind); % Start Timer tic % Load input parameter file inputParameters; % Measurement comments to add to Filename % (Note: Files are guaranteed to be unique by date and time stamp) prompt = {'Insert comment about measurement location'}; dlg_title = 'Measurement Setup and Location:'; num_lines = 1; def = {''};  89  option.Resize = 'on'; option.WindowStyle = 'modal'; option.Interpreter = 'tex'; response = inputdlg(prompt, dlg_title, num_lines, def, option); comment = [comment cell2mat(response)]; % Initialize Positioner disp(sprintf('Initializing Linear Positioner:')); positionerInit(comPort, baudrate, positionerTimeout, ... stepSize, xMotorGuard, yMotorGuard); % Initialize PNA [pna] = PNAinit(vendor, ... boardNumber, deviceNumber, PNAInputBufferSize, ... PNATimeout, numPoints, start, stop, ... ifBandwidth, rfPower); disp(sprintf('PNA intialization complete.\r')); % Acquire Data and plot disp(sprintf('Starting DCS measurements:\r')); startMeasurement; % Close PNA PNAclose(pna); disp(sprintf('PNA communication closed.\r')); % Close Positioner positionerClose(positionerTimeout); % End Timer measDuration = toc; disp(sprintf('Measurement time was %0.7f seconds.\r', measDuration)); % Post-Processing fileName = generateFileName(nX, nY, start, stop, comment); if (strcmp(postProcEnable, 'Y')); [Hf, frequency] = processRawData(measuredData, calData, ... returnLoss_UWB_823, returnLoss_UWB_765, antennaCalData, ... start, stop); subplot(2,1,1); plot(frequency, 20*log10(abs(reshape(Hf(1,1,1,1:numPoints), ... [1 numPoints])))); title('Plot of Measured Amplitude'); xlabel('Frequency (GHz)'); ylabel('Amplitude (dB)'); axis([start/(1E9) stop/(1E9) magMin magMax]); subplot(2,1,2); plot(frequency, (angle(reshape(Hf(1,1,1,1:numPoints), ... [1 numPoints]))*180/pi)); title('Plot of Measured Phase'); xlabel('Frequency (GHz)') ylabel('Phase (deg)'); axis([start/(1E9) stop/(1E9) -180 180]); saveas(gcf,strcat(fileName, ' - CFR.jpg') ,'jpeg'); [pdp, htau, t] = cfr2cir(measuredData, calData, returnLoss_UWB_823, ... returnLoss_UWB_765, antennaCalData, start, stop); apdp = avgPDP(pdp); apdp = apdp./max(apdp); % Set maximum to 0dB PDPplot(apdp, t, c0,tMax); saveas(gcf,strcat(fileName, ' - CIR.jpg') ,'jpeg'); % Calculate Distance distance = t(find(10*log10(apdp) == 0, ...  90  1, 'first'))*c0; % Find the maximum Peak disp(sprintf('Calculated antenna seperation is %1.3f m\r', ... distance)); end; % Save Data save(strcat(fileName, '.mat')); disp(sprintf('Saved acquired data to file "%s"\r', fileName)); % End Program disp(sprintf('UWB Channel Sounder measurements complete'))  inputParameters.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Directional Channel Sounder Parameter File % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - June 1, 2008 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Choose measurement type measurementType = 4; % Default to UWB Mode if (measurementType == 1) % 433 MHz measurement parameters comment = '433 Mhz - '; % PNA Measurement Paramters numPoints = 1601; % Number of Points start = 410E6; % Start Frequency stop = 470E6; % Stop Frequency ifBandwidth = 1000; % IF Bandwidth rfPower = 0; % Transmit Power in dBm (b/w -25dBm to 0dBm) % Linear Positioner Parameters nX = 3; % # of x-axis points on virtual array nY = 2; % # of y-axis points on virtual array c0 = 299792458; % speed of light in a vacuum spacingFreq = 433E6; % Default Frequency is 433 MHz % Load Calibration Data calDataFileName = 'calData_433M'; load(calDataFileName); % Load calibration Data elseif (measurementType == 2) % 868 MHz measurement parameters comment = '868 Mhz - '; % PNA Measurement Paramters numPoints = 1601; % Number of Points start = 838E6; % Start Frequency stop = 898E6; % Stop Frequency ifBandwidth = 1000; % IF Bandwidth rfPower = 10; % Transmit Power in dBm (b/w -25dBm to 0dBm) % Linear Positioner Parameters nX = 4; % # of x-axis points on virtual array nY = 3; % # of y-axis points on virtual array c0 = 299792458; % speed of light in a vacuum spacingFreq = 868E6; % Default Frequency is 868 MHz % Load Calibration Data calDataFileName = 'calData_868M'; load(calDataFileName); % Load calibration Data elseif (measurementType == 3) % 2450 MHz measurement parameters comment = '2450 Mhz - ';  91  % PNA Measurement Paramters % Number of Points numPoints = 1601; start = 2410E6; % Start Frequency stop = 2490E6; % Stop Frequency % IF Bandwidth ifBandwidth = 1000; rfPower = 10; % Transmit Power in dBm (b/w -25dBm to 0dBm) % Linear Positioner Parameters nX = 6; % # of x-axis points on virtual array nY = 6; % # of y-axis points on virtual array % speed of light in a vacuum c0 = 299792458; spacingFreq = 2450E6; % Default Frequency is 2450 MHz % Load Calibration Data calDataFileName = 'calData_2450M'; load(calDataFileName); % Load calibration Data elseif (measurementType == 4) % UWB measurement parameters comment = 'UWB - '; % PNA Measurement Paramters % Number of Points numPoints = 6401; start = 3100E6; % Start Frequency stop = 10600E6; % Stop Frequency % IF Bandwidth ifBandwidth = 3000; rfPower = -20; % Transmit Power in dBm (between -25dBm to 0dBm) % Linear Positioner Parameters nX = 7; % # of x-axis points on virtual array nY = 7; % # of y-axis points on virtual array c0 = 299792458; % speed of light in a vacuum spacingFreq = 3100E6; % Default Frequency is 3100 MHz % Load Calibration Data calDataFileName = 'calData_UWB'; load(calDataFileName); % Load calibration Data % Load Return Loss Corrections load('ReturnLossUWB.mat'); % Load Antenna Corrections load('antennaUWB_90deg.mat'); %load('antennaUWB_80_100deg.mat'); %antennaCalData = ones(6401, 1); end; % Linear Positioner Spacing Parameters spacingXLambda = 0.5; % x-axis delta is 0.5*lambda spacingYLambda = 0.5; % y-axis delta is 0.5*lambda deltaX = c0/(spacingFreq)*spacingXLambda*1E3; % X-axis spacing (in mm) deltaY = c0/(spacingFreq)*spacingYLambda*1E3; % Y-axis spacing (in mm) % MATLAB Figure Plotting Parameters plotEnable = 'N'; % Enable Plotting tMax = 100*(1E-9); % Maximum time for PDP plots magMax = -30; % in dB magMin = -110; % in dB % Temporal Averaging Parameters temporalAvgEnable = 'N'; % Enable Temporal Averaging temporalAvgSamples = 1; % Number of traces to acquire per location temporalAvgDelay = 5; % Delay in seconds between acquisitions % Post Processing Parameters postProcEnable = 'Y';  % Enable Post Processing 'Y' or 'N' % CFR2CIR, APDP, ClusterID, etc.  % PNA GPIB Connection Parameters vendor = 'ni'; % GPIB vendor boardNumber = 0; % Board Number of GPIB board  92  deviceNumber = 16; PNAInputBufferSize = 2^20; PNATimeout = 60; % Positioner Parameters stepSize = 0.005; xMotorGuard = 3; yMotorGuard = 3;  % GPIB address of PNA % Input memory buffer for data download % PNA inactivity timeout in seconds  % Step size is 0.005mm % 3mm Guard from Limit Switch  % Positioner Conenction Parameters comPort = 5; baudrate = 9600; positionerTimeout = 20000; % in milliseconds (10 sec)  positionerInit.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Velmex 2D Linear Positioner % % Initialization % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - Aug 23, 2007 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function positionerInit(comPort, baudrate, timeout, ... stepSize, xMotorGuard, yMotorGuard) % Load Velmex VMX Controller MATLAB Drivers LoadDriver; % Open Communication with the VMX System PortClose; % Close any previous ports portOpenError = PortOpen(comPort, baudrate); if portOpenError == 1 disp(sprintf(' - Communication with VMX Controller successfully established')); else ReleaseDriver; error('Error establishing communication with VMX Controller'); end; PortSendCommands('F,C,'); % Disable echo Mode % Setup Motors and Limit Switches PortSendCommands('setM1M5,'); % Motor 1 (Vexta PK268-3A) is along the Xaxis (1m positioner) %PortSendCommands('setM1M4,'); % Motor 1 (Vexta PK266-3A) is along the X-axis (0.5m positioner) PortSendCommands('setM2M4,'); % Motor 2 (Vexta PK266-3A) is along the y-axis PortSendCommands('setL1M1,'); % Enable Motor 1 Limit Switches to N/C PortSendCommands('setL2M1,'); % Enable Motor 2 Limit Switches to N/C PortSendCommands('R'); % Run Program portSendError = PortWaitForChar('^', timeout); % Wait for program to finish if portSendError == 1 disp(sprintf(' - Motors and Limit Switches configured successfully.')); else PortClose; ReleaseDriver; error(sprintf('Error configuring Motors \r')); end; % % Move DCS to Origin (0,0) % Move X-Axis  93  disp(' - Moving X-Axis to origin...'); command = strcat('C,S1M1500,I1M-1000,I1M0,I1M',num2str(xMotorGuard/stepSize),',IA1M-0,R'); PortSendCommands(command); portSendError = PortWaitForChar('^', 36*timeout); finish if portSendError == 1 disp(sprintf('\bdone!')); else PortClose; ReleaseDriver; error('Error moving X-Axis to origin'); end; % Move Y-Axis disp(' - Moving Y-Axis to origin...'); command = strcat('C,S2M1500,I2M-1000,I2M0,I2M',num2str(yMotorGuard/stepSize),',IA2M-0,R'); PortSendCommands(command); portSendError = PortWaitForChar('^', 18*timeout); finish if portSendError == 1 disp(sprintf('\bdone!\r')); else PortClose; ReleaseDriver error(sprintf('Error moving Y-Axis to origin \r')); end;  % Wait for program to  % Wait for program to  PNAInit.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Agilent PNA E8362B Series Network Analyzer % % Initialization % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - Aug 23, 2007 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function [pna] = PNAinit(vendor, boardNumber, deviceNumber, ... PNAInputBufferSize, PNATimeout, numPoints, start, ... stop, ifBandwidth, rfPower) % Set up Dwell Time dwellTime = 0.0002;  % Set Dwell Time to 200us  % Set up the connection with PNA pna = gpib(vendor, boardNumber, deviceNumber); % Configure VNA buffer size and timeout set(pna, 'InputBufferSize', PNAInputBufferSize); set(pna, 'Timeout', PNATimeout); fopen(pna); % PNA Reset and initialization fprintf(pna, 'SYST:FPR'); % Factory presets fprintf(pna, 'CALC:PAR:DEF My_S21, br1, 1'); % Create a measurement fprintf(pna, 'DISP:ENAB OFF'); % Turn on/off display fprintf(pna, 'INIT:CONT OFF'); % Set for manual sweep fprintf(pna, '*OPC?'); opc = fscanf(pna);  % wait until init complete % read output to clear queue  94  fprintf(pna, 'SENS:CORR OFF'); % Disable Calibration %fprintf(pna, 'SENS:CORR:CSET:ACT "CAL_dcs",1'); % Set Calibration to use % Configure Measurement fprintf(pna, 'SENS:SWE:GEN STEP'); fprintf(pna, 'SENS:SWE:DWEL %f;', dwellTime'); fprintf(pna, 'SENS:SWE:TYPE LIN'); fprintf(pna, 'SENS:SWE:TRIG:POIN OFF'); fprintf(pna, 'SENS:SWE:POIN %.0f;', numPoints); fprintf(pna, 'SENS:FREQ:STAR %.0f;', start); fprintf(pna, 'SENS:FREQ:STOP %.0f;', stop); fprintf(pna, 'SENS:BWID %.0f;', ifBandwidth); fprintf(pna, 'SOUR:POW:ATT:AUTO OFF'); fprintf(pna, 'SOUR:POW %.0f;', rfPower);  % % % % % % % % % %  % Get Sweep Time %fprintf(pna, 'SENS:SWE:TIME?'); %fscanf(pna)  % Query Sweep Time % read back sweep time  % Get Generation Type %fprintf(pna, 'SENS:SWE:GEN?'); %fscanf(pna)  % Query Sweep Time % read back sweep time  %fprintf(pna, 'CALC:CORR:EDEL:TIME -95e-12');  % Set electrical delay  Set Sweep Generation to Stepped Set Dwell Time Linear Frequency Measurement measure all points/trigger number of points start frequency stop frequency IF Bandwidth Disable Source Attenuators Set RF Output Power  startMeasurement.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Start Measuremennt % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - June 1, 2008 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Display Measurements Parameters disp(sprintf('Measurement Parameters:')); disp(sprintf(' - %3.0f MHz to %3.0f MHz with %.0f points', ... (start/1E6), (stop/1E6), numPoints)); disp(sprintf(' - %d by %d array', nX, nY)); disp(sprintf(' - The x-axis spacing is %3.3fmm', deltaX)); disp(sprintf(' - The y-axis spacing is %3.3fmm', deltaY)); if (strcmp(temporalAvgEnable, 'Y')); disp(sprintf(' - Temporal Sampling Enabled:')); disp(sprintf(' - %d samples per location', temporalAvgSamples)); disp(sprintf(' - %d second delay between acquisitions\r', ... temporalAvgDelay)); else disp(sprintf(' - Temporal Sampling Disabled\r')); end; % Declare variable for magnitude and phase if (strcmp(temporalAvgEnable, 'Y')) measuredData = zeros([nX nY temporalAvgSamples numPoints]); else measuredData = zeros([nX nY 1 numPoints]); end; %Move Antenna and Take measurements reverse = 0; % Take measurements at (0,0) location [measuredData(1,1,:,:)] = acquireData(pna, 1, 1, plotEnable, ... temporalAvgEnable, temporalAvgSamples, temporalAvgDelay, ... start, stop, magMin, magMax);  95  % Take measurements at (0,nY) locations if (nY > 1) for (b = 2:nY) command = strcat('C,S2M6000,I2M-',int2str(deltaY/stepSize),',R'); PortSendCommands(command); portSendError = PortWaitForChar('^', positionerTimeout); % Wait for program to finish if portSendError == 1 [measuredData(1,b,:,:)] = acquireData(pna, 1, b, plotEnable, ... temporalAvgEnable, temporalAvgSamples, temporalAvgDelay, ... start, stop, magMin, magMax); else PortClose; ReleaseDriver; error('Error error while moving along the y-axis'); end reverse = 1; end end if (nX > 1) for (a = 2:(nX)) command = strcat('C,S1M6000,I1M-',int2str(deltaX/stepSize),',R'); PortSendCommands(command); % Wait for portSendError = PortWaitForChar('^', positionerTimeout); program to finish if (portSendError == 1) if (reverse == 1); [measuredData(a,nY,:,:)] = acquireData(pna, a, nY, plotEnable, ... temporalAvgEnable, temporalAvgSamples, temporalAvgDelay, ... start, stop, magMin, magMax); elseif (reverse == 0) [measuredData(a,1,:,:)] = acquireData(pna, a, 1, plotEnable, ... temporalAvgEnable, temporalAvgSamples, temporalAvgDelay, ... start, stop, magMin, magMax); end if (nY > 1) for (b = 2:(nY)) if (reverse == 1) command = strcat('C,S2M6000,I2M',int2str(deltaY/stepSize),',R'); PortSendCommands(command); % portSendError = PortWaitForChar('^', positionerTimeout); Wait for program to finish if portSendError == 1 [measuredData(a,nY-b+1,:,:)] = acquireData(pna, a, nY-b+1, ... plotEnable, temporalAvgEnable, temporalAvgSamples, ... temporalAvgDelay, start, stop, magMin, magMax); else PortClose; ReleaseDriver error('Error error while moving along the y-axis'); end elseif (reverse == 0) command = strcat('C,S2M6000,I2M-',int2str(deltaY/stepSize),',R'); PortSendCommands(command); portSendError = PortWaitForChar('^', positionerTimeout); % Wait for program to finish if portSendError == 1 [measuredData(a,b,:,:)] = acquireData(pna, ... a, b, plotEnable, temporalAvgEnable, ... temporalAvgSamples, temporalAvgDelay, ... start, stop, magMin, magMax); else PortClose; ReleaseDriver  96  error('Error error while moving along the y-axis'); end else error('Unknown Program Error'); end end end reverse = bitcmp(reverse, 1); else PortClose; ReleaseDriver error('Error while moving along the x-axis'); end end end function [data] = acquireData(pna, a, b, plotEnable,... AvgEnable, AvgSamples, AvgDelay, ... start, stop, magMin, magMax) disp(sprintf('Antenna at position (%d,%d). Acquiring data...', ... (a-1), (b-1))); % Take Measurement with PNA if (strcmp(AvgEnable, 'Y')) pause on; for (i = 1:AvgSamples) disp(sprintf(' Starting time sample %d...', i)); [data(1,1,i,:)] = PNAacquire(pna); disp(sprintf('\bdone!')); if (i ~= AvgSamples) pause(AvgDelay); end; end; disp(sprintf(' Location done!\r')); pause off; else [data(1,1,1,:)] = PNAacquire(pna); disp(sprintf('\bdone!\r')); end; % Plot Data for single location if (strcmp(plotEnable, 'Y')) PNAplot(data(1,1,:,:), start, stop, magMin, magMax); end;  processRawData.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Process Raw Measurement Data % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - June 1, 2008 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function [Hf, frequency] = processRawData(data, calData, returnLoss_UWB_Tx, ... returnLoss_UWB_Rx, antennaCalData, start, stop) % Get Measurement Matrix dimensions [nX nY AvgSamples numPoints] = size(data); % Calculate Frequency frequency = (start:((stop-start)/(numPoints-1)):stop)./(1E9);  97  % Add thru-line calibration to CFR H(f) measurement cal = repmat(reshape(calData, [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = data./cal; % Apply Frequency Response Correction % Add Tx Return Loss calibration to CFR H(f) measurement transmissionLossTx = sqrt(1-abs(returnLoss_UWB_Tx).^2); transmissionLossTx = repmat(reshape(transmissionLossTx, ... [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = Hf.*transmissionLossTx; % Apply Tx Return Loss Correction % Add Rx Return Loss calibration to CFR H(f) measurement transmissionLossRx = sqrt(1-abs(returnLoss_UWB_Rx).^2); transmissionLossRx = repmat(reshape(transmissionLossRx, ... [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = Hf.*transmissionLossRx; % Apply Rx Return Loss Correction % Add UWB Antenna Calibration (compenstate for non-linear omnidirectionality) antennaCal = repmat(reshape(antennaCalData, [1 1 1 numPoints]), ... [nX nY AvgSamples]); Hf = Hf./antennaCal; % Apply UWB Antenna Calibration %Calculate average magnitude AvgHf = mean(mean(Hf));  % Average Frequency Response  cfr2cir.m %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Convert CFR to CIR % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% % Created by: Adam Stefanski - June 1, 2008 % %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% function [pdp, htau_round, t] = cfr2cir(data, calData, returnLoss_UWB_Tx, ... returnLoss_UWB_Rx, antennaCalData, start, stop) % Get Measurement Matrix dimensions [nX nY AvgSamples numPoints] = size(data); % Create Time Variable t = (0:1/(stop-start):floor(numPoints/2)/(stop-start)); % Add thru-line calibration to CFR H(f) measurement cal = repmat(reshape(calData, [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = data./cal; % Apply Frequency Response Correction % Add Tx Return Loss calibration to CFR H(f) measurement transmissionLossTx = sqrt(1-abs(returnLoss_UWB_Tx).^2); transmissionLossTx = repmat(reshape(transmissionLossTx, ... [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = Hf.*transmissionLossTx; % Apply Tx Return Loss Correction % Add Rx Return Loss calibration to CFR H(f) measurement transmissionLossRx = sqrt(1-abs(returnLoss_UWB_Rx).^2); transmissionLossRx = repmat(reshape(transmissionLossRx, ... [1 1 1 numPoints]), [nX nY AvgSamples]); Hf = Hf.*transmissionLossRx; % Apply Rx Return Loss Correction % Add UWB Antenna Calibration (compenstate for non-linear omnidirectionality) antennaCal = repmat(reshape(antennaCalData.^2, ... [1 1 1 numPoints]), [nX nY AvgSamples]); % Apply UWB Antenna Calibration Hf = Hf./antennaCal;  % Apply Windowing %Hf = Hf./ATF; %Wf = hann(numPoints); %Wf = hamming(numPoints); Wf = kaiser(numPoints, 7); %Wf = ones(numPoints);  % % % %  Apply Use a Use a Use a  Frequency Response Correction Hanning Window per Karadel 2007 Hamming Window per CCC 2005 Kaiser Window per Chuang 2008  98  % % Change CFR H(f) to CIR h(tau) % for (x = 1:nX) % for (y = 1:nY) % for (z = 1:AvgSamples) % temporal averaging % htau(x,y,z,:) = ifft(Hf(x,y,z,:) .* ... % reshape(Wf, [1 1 1 numPoints])); % htau_round(x,y,z,:) = htau(x,y,z,(1:ceil(numPoints/2))); % htau_round(x,y,z,:) = ... % htau_round(x,y,z,:)/sqrt(sum(abs(htau_round(x,y,z,:)).^2)); % end; % end; % end; % Calculate Channel Impulse Response (Hermitian Approach) Hhat = zeros([nX nY AvgSamples (2*ceil(stop/((stop-start)/(numPoints - 1)))1)]); for (x = 1:nX) for (y = 1:nY) for (z = 1:AvgSamples) Pf = Hf(x,y,z,:) .* reshape(Wf, [1 1 1 numPoints]); Hhat(x,y,z,(1:1:numPoints)) = conj(Pf(1,1,1,(end:-1:1))); Hhat(x,y,z,(length(Hhat)-length(Pf)+1):1:length(Hhat)) = Pf; htau(x,y,z,:) = ifft(Hhat(x,y,z,:)); htau(x,y,z,:) = htau(x,y,z,:)/sqrt(sum(abs(htau(x,y,z,:)).^2)); clear Pf; end; end; end; t = (0:1/(2*stop):(length(htau)-1)/(2*stop)); htau_round = htau; % Calculate Power Delay Profile pdp = abs(htau_round).^2;  99  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

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"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            data-media="{[{embed.selectedMedia}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
https://iiif.library.ubc.ca/presentation/dsp.24.1-0064889/manifest

Comment

Related Items