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

Deployment of wireless sensor and actuator networks for precision agriculture Sivertsen, Kyle N. 2013

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

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

Item Metadata

Download

Media
24-ubc_2013_fall_sivertsen_kyle.pdf [ 1.95MB ]
Metadata
JSON: 24-1.0074265.json
JSON-LD: 24-1.0074265-ld.json
RDF/XML (Pretty): 24-1.0074265-rdf.xml
RDF/JSON: 24-1.0074265-rdf.json
Turtle: 24-1.0074265-turtle.txt
N-Triples: 24-1.0074265-rdf-ntriples.txt
Original Record: 24-1.0074265-source.json
Full Text
24-1.0074265-fulltext.txt
Citation
24-1.0074265.ris

Full Text

Deployment of Wireless Sensor and Actuator Networks for Precision Agriculture  by  Kyle N. Sivertsen  B.A.Sc., The University of British Columbia, 2007  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF APPLIED SCIENCE  in  THE FACULTY OF GRADUATE AND POSTDOCTORAL STUDIES  (Electrical and Computer Engineering)  THE UNIVERSITY OF BRITISH COLUMBIA  (Vancouver)     August 2013  ? Kyle N. Sivertsen, 2013  ii Abstract  Wireless sensor and actuator networks (WSAN) are emerging as a key enabling technology for precision agriculture, a technique for maximizing crop yield and quality through targeted application of resources such as water, fertilizer and pest control agents by exploiting temporal and spatial variability in crop and soil conditions. In this thesis, we make three contributions to the field. First, we assess the state of the art in deployment and configuration of wireless sensor and actuator networks for precision agriculture, including the relevance and suitability of existing propagation models, lessons learned from previous demonstrations and field trials, and the potential for improving network performance through suitable deployment strategies, and physical, medium access control (MAC) and network layer design. We reveal an urgent need to assess airlink design of such networks to account for the unique nature of the wireless propagation environment and to consolidate proposed improvements to and best practices for WSAN design in the form of an industry standard. Second, we show that the conventional practice of employing wireless transceivers  that operate at 800 MHz or above incurs significant penalties for achievable range and/or power consumption and propose that low-power short-range wireless devices intended for use as sensor nodes in precision agriculture be allowed to share the 433 MHz sub-band currently authorized for use by active radio frequency identification (RFID) devices at cargo terminals, port facilities and warehouses so that they may experience less path loss and achieve greater range and reliability while consuming less power. Finally, we analyze 2450 MHz channel impulse responses that we measured in a high-density apple orchard and consider the implications of both their form and scale for the design and deployment of WSANs and our understanding of the propagation environment. Of particular note is the vastly reduced delay spread compared to that observed in traditional residential, commercial and industrial environments.  iii Preface  This thesis is original, unpublished, independent work by the author, Kyle N. Sivertsen, and was conducted under the supervision of Prof. David G. Michelson. Section 2.5 of the literature survey presented in Chapter 2 was conducted in collaboration with Pooyan Abouzar.   ivTable of Contents  Abstract .................................................................................................................................... ii Preface ..................................................................................................................................... iii Table of Contents ................................................................................................................... iv List of Tables ........................................................................................................................ viii List of Figures ......................................................................................................................... ix List of Abbreviations ............................................................................................................. xi Acknowledgements .............................................................................................................. xiii Dedication ............................................................................................................................. xiv Chapter 1: Introduction ......................................................................................................... 1 Chapter 2: Deployment and Configuration of Wireless Sensor and Actuator Networks for Precision Agriculture ........................................................................................................ 3 2.1 Introduction ........................................................................................................................... 3 2.1.1 Previous Surveys of PA and WSAN ................................................................................ 3 2.1.2 Limitations of Previous Work .......................................................................................... 7 2.1.3 Objectives ......................................................................................................................... 8 2.2 Emergence of Precision Agriculture ..................................................................................... 9 2.2.1 Current Challenges in Agriculture .................................................................................... 9 2.2.2 Limitations of Traditional Solutions ............................................................................... 10 2.2.3 Precision Agriculture ...................................................................................................... 10 2.2.4 Important Milestones in the Development of PA ........................................................... 12 2.2.5 Current Challenges in PA ............................................................................................... 13 2.3 Wireless Propagation in Vegetated Environments: Literature Survey ................................ 14 2.3.1 Physical Nature of the Agricultural Environment .......................................................... 14 2.3.2 Characterization and Modeling in Other Fields of Study ............................................... 19 2.4 Demonstrations and Field Trials of WSAN in PA Applications ........................................ 19 2.4.1 Role of Demonstrations and Field Trials ........................................................................ 20 2.4.2 Agricultural Applications ............................................................................................... 21 2.4.3 Environments .................................................................................................................. 25  v2.4.4 Wireless Technologies .................................................................................................... 27 2.4.5 Growth and Sophistication ............................................................................................. 28 2.4.6 Lessons Learned ............................................................................................................. 31 2.5 Performance Issues and Enabling Technologies ................................................................. 32 2.5.1 Node Placement and Network Topology ........................................................................ 33 2.5.2 Data Compression and Node Management .................................................................... 33 2.5.3 MAC Layer Techniques ................................................................................................. 34 2.5.4 Network Layer ................................................................................................................ 35 2.5.5 Potential for Cross-Layer Design ................................................................................... 37 Chapter 3: Spectrum Allocation for Short Range Wireless Devices in Precision Agriculture............................................................................................................................. 39 3.1 Introduction ......................................................................................................................... 39 3.1.1 Significance of Spectrum Selection ................................................................................ 39 3.1.2 Trends in Wireless .......................................................................................................... 40 3.1.3 Objectives ....................................................................................................................... 41 3.2 Requirements and Challenges ............................................................................................. 41 3.2.1 Environment ................................................................................................................... 41 3.2.2 Power Consumption........................................................................................................ 42 3.2.3 Range: Node Spacing and Sampling Density ................................................................. 42 3.3 Previous Work and its Limitations ...................................................................................... 44 3.3.1 Recent Allocation Efforts ............................................................................................... 44 3.3.1.1 Opening the 433 MHz Sub-band for RFID ............................................................ 44 3.3.1.2 DSRC Allocations at 5.8 GHz in the Late 1990?s ................................................. 45 3.3.1.3 ITS Allocations at 5 GHz in Europe ...................................................................... 47 3.3.1.4 Digital Dividend Band for Mobile Communications in Europe ............................ 48 3.3.1.5 Harmonization Analysis for Wireless Medical Applications ................................. 48 3.3.2 Carrier Frequency and System Guidelines for Agriculture and WSAN ......................... 49 3.3.3 Carrier Frequency and System Guidelines for Ubiquitous Sensor Networks ................. 50 3.3.4 Models of Path Loss in Vegetated Environments ........................................................... 52 3.3.4.1 Modified Exponential Decay (MED) ..................................................................... 53 3.3.4.2 Modified Gradient Models ..................................................................................... 55 3.3.4.3 Ground and Canopy Reflections ............................................................................ 56 3.3.4.4 Log-Distance Model .............................................................................................. 57  vi3.3.4.5 Radiative Energy Transfer (RET) Theory.............................................................. 58 3.3.4.6 Comparisons and Goodness of Fit ......................................................................... 58 3.3.4.7 Discussion .............................................................................................................. 60 3.4 Key Issues for Spectrum Selection ..................................................................................... 61 3.4.1 Harmonization ................................................................................................................ 61 3.4.2 Path Loss and Propagation Impairments ........................................................................ 61 3.4.2.1 Range Comparison ................................................................................................. 61 3.4.3 Spectrum Occupancy ...................................................................................................... 68 3.4.4 Interference Avoidance and Mitigation .......................................................................... 68 3.5 Potential Bands ................................................................................................................... 68 3.5.1 List of ISM Bands:.......................................................................................................... 68 3.5.2 The 2.4 GHz Band .......................................................................................................... 69 3.5.3 The 915 and 868 MHz Bands ......................................................................................... 69 3.5.4 The 433 MHz Band ........................................................................................................ 70 3.6 Discussion ........................................................................................................................... 72 Chapter 4: Interpretation and Implications of the Channel Impulse Response Observed at 2450 MHz in High Density Apple Orchards .................................................................. 76 4.1 Introduction ......................................................................................................................... 76 4.1.1 Significance of Propagation Environment and Delay Spread ......................................... 76 4.1.2 Previous Work ................................................................................................................ 77 4.1.3 Limitations of Previous Work ........................................................................................ 80 4.1.4 Objectives ....................................................................................................................... 80 4.1.5 Outline ............................................................................................................................ 81 4.2 Propagation Environment ................................................................................................... 81 4.2.1 Details of the Environment Studied ................................................................................ 81 4.2.2 Propagation Mechanisms ................................................................................................ 82 4.2.3 Principle of Measurement ............................................................................................... 84 4.2.4 Shape of Impulse Response ............................................................................................ 85 4.3 Measurement Configuration and Procedure ........................................................................ 86 4.3.1 Equipment ....................................................................................................................... 86 4.3.2 Calibration and Validation .............................................................................................. 88 4.3.3 Measurement Configurations .......................................................................................... 89 4.4 Results and Reductions ....................................................................................................... 90  vii4.4.1 Data Processing .............................................................................................................. 90 4.4.2 RMS Delay Spread ......................................................................................................... 95 4.4.2.1 Moment-Method .................................................................................................... 95 4.4.2.2 Slope-Method ......................................................................................................... 97 4.4.3 Coherence Bandwidth ................................................................................................... 100 4.4.4 Path Loss ....................................................................................................................... 101 4.4.5 RMS Delay Spread and Distance.................................................................................. 103 4.4.6 RMS Delay Spread and Path Loss ................................................................................ 106 4.4.7 Coherence Bandwidth and RMS Delay Spread. ........................................................... 108 4.5 Discussion and Implications ............................................................................................. 110 4.5.1 Interpretation ................................................................................................................ 110 4.5.2 Design ........................................................................................................................... 111 4.5.3 Deployment .................................................................................................................. 111 Chapter 5: Conclusions and Recommendations .............................................................. 113 References ............................................................................................................................ 119 References ............................................................................................................................ 119   viiiList of Tables  Table 1 ? Path loss models, parameters, and range predictions a 433, 915, and 2400 MHz .. 67 Table 2 ? Measurement configurations ................................................................................... 90   ixList of Figures  Figure 1 ? Proportion of demonstrations and field trials which have been conducted for each agricultural application ........................................................................................................... 22 Figure 2 ? Proportion of deployments in each environment ................................................... 25 Figure 3 ? Wireless technologies used by nodes .................................................................... 28 Figure 4 ? Wireless technologies used by backhaul ............................................................... 28 Figure 5 ? Sophistication of trial WSAN realizations ............................................................ 29 Figure 6 ? Sophistication of trial WSAN realizations over time ............................................ 29 Figure 7 ? Sophistication of WSAN analysis ......................................................................... 30 Figure 8 ? Sophistication of WSAN analysis over time ......................................................... 30 Figure 9 ? Maximum communications range for -80, -90, and -100 dBm RX sensitivity ..... 63 Figure 10 ? Maximum communications range for -80, -90, and -100 dBm RX sensitivity (shorter y-axis) ........................................................................................................................ 63 Figure 11 ? (a) ?High-vertical Rows? consist of single apple tree branches planted at regular intervals. (b) ?V-shaped rows? consist of two branches per row planted at an angle to maximize tree density. ............................................................................................................ 82 Figure 12 ? (a) Multipath components in an apple orchard. (b) Interaction between the layers of the propagation environment. ............................................................................................. 83 Figure 13 ? ?Spike plus exponential? delay profile shape ...................................................... 86 Figure 14 ? Measurement Configuration ................................................................................ 87 Figure 15 ? Propagation Directions ........................................................................................ 90 Figure 16 ? A single snapshot of the channel frequency response (magnitude and phase) for along-row propagation, high TX antenna height, and 6.3m antenna separation. ................... 91 Figure 17 ? Un-calibrated PDP for along-row propagation, high TX antenna height, and 6.3m antenna separation. .................................................................................................................. 93 Figure 18 ? Calibrated PDP for along-row propagation, high TX antenna height, and 6.3m antenna separation. .................................................................................................................. 94 Figure 19 ? Processed power delay profile for along-row propagation, high TX antenna height, and 6.3m antenna separation. ...................................................................................... 95 Figure 20 ? The effect of threshold selection on estimates of RMS delay spread .................. 97  xFigure 21 ? ?Spike + Exponential? fit to power-delay profile for along-row propagation, high  TX antenna height, and 6.3 m antenna separation. ................................................................. 99 Figure 22 - ?Spike + Exponential? fit to power-delay profile for along-row propagation, low  TX antenna height, and 100 m antenna separation. ................................................................ 99 Figure 23 ? Coherence bandwidth for correlation level of 0.9 for along-row propagation, low  TX antenna height, and 6.3 m antenna separation. ............................................................... 101 Figure 24 ? Path loss vs. antenna separation and linear regression ...................................... 103 Figure 25 ? Moment-method RMS delay spread estimates vs. antenna separation. ............ 105 Figure 26 ? Slope-method RMS delay spread estimates vs. antenna separation. ................. 105 Figure 27 ? Moment-method RMS delay spread vs. path loss. ............................................ 107 Figure 28 ? Slope-method RMS delay spread vs. path loss. ................................................ 108 Figure 29 ? Coherence bandwidth vs. moment-method RMS delay spread......................... 109 Figure 30 ? Coherence bandwidth vs. slope-method RMS delay spread ............................. 109     xiList of Abbreviations  WSAN Wireless Sensor and Actuator Network WSN  Wireless Sensor Network MAC  Medium Access Control RFID  Radio Frequency Identification COTS  Commercial Off The Shelf ISM  Industrial Scientific and Medical SAR  Synthetic Aperture Radar PA  Precision Agriculture SSCM  Site-Specific Crop Management RET  Radiative Energy Transfer  VRT  Vector Radiative Transfer  SNR  Signal-to-Noise Ratio  BER  Bit Error Rate DSS  Decision Support System GSM  Global Systems for Mobile communications GPRS  General Packet Radio Service TVWS  Television Whitespace LTE  Long Term Evolution VHF  Very High Frequency UHF  Ultra High Frequency SRD  Short Range Devices RX  Receiver TX  Transmitter LOS  Line Of Site RMS  Root Mean Square RF  Radio Frequency VNA  Vector Network Analyzer PVC  Polyvinyl Chloride SCPI  Standard Commands for Programmable Instruments  xiiVISA  Virtual Instrument Software Architecture CRF  Channel Frequency Response CIR  Channel Impulse Response PDP  Power Delay Profile CBW  Coherence Bandwidth    xiiiAcknowledgements  I am sincerely grateful to the faculty and staff at UBC whose teaching and advice I have benefited from over the course of my education. I am particularly grateful to my fellow students in the Radio Science Lab who have helped me immeasurably and to whom I owe a great debt.   I would specifically like to thank to Dr. David G. Michelson for all of his countless hours of advice, and teaching. I have learned many important lessons from him about research, hard work, and achieving success. I have no doubt that my career will be forever impacted by his tutelage.   My family has provided me with endless support over many difficult years and for that I will forever be grateful. I would never have been able to complete my work without them.   xiv Dedication  This thesis is dedicated to my wife, Anna Sivertsen. I will never be able to repay her for the support and understanding she has shown me during this time.   1Chapter 1: Introduction  Wireless systems operating in the presence of dense vegetation experience channel impairment due to the absorption and scattering caused by foliage in the path of propagating signals. The severity of these impairments increases when the terminals of such systems are placed within the vegetation canopy, as is often the case for deployments of wireless networks in agricultural environments. Interest in deploying wireless systems in agriculture has increased in recent years as growers seek to monitor and control sensors and actuators to assist a variety of agricultural processes. These wireless sensor and actuator networks (WSNs or WSANs) employ ad-hoc, mesh networking to connect individual nodes which are deployed within fields, orchards, and plantations, and thus far from convenient sources of power. Such nodes therefore have strict power consumption requirements, as they must reliably monitor and control plant, soil, climate, environment, and pest conditions over the length of the growing season. Commercial off-the-shelf (COTS) hardware and software has been available for WSAN for some time and has been used in (mostly indoor) commercial and industrial environments. Although there has been considerable interest on the part of government, university, and corporate researchers in the potential of agricultural WSANs to improve production while reducing waste, environmental impact, and use of scarce resources, little work has been done to investigate whether existing WSAN frequency bands, network technologies, and devices are appropriate for agricultural environments. In this thesis, we make three contributions to the field. In Chapter Two, we assess the state of the art in deployment and configuration of wireless sensor and actuator networks for precision agriculture, including the relevance and suitability of existing propagation models, lessons learned from previous demonstrations and field trials, and the potential for improving network performance through suitable deployment strategies, and physical, medium access control (MAC) and network layer design. We reveal an urgent need to assess airlink design of such networks at all layers to account for the unique nature of the wireless propagation environment and to consolidate proposed improvements to and best practices for WSAN design in the form of an industry standard.   2In Chapter Three, we show that the conventional practice of employing wireless transceivers  that operate at 800 MHz or above incurs significant penalties for achievable range and/or power consumption and propose that low-power short-range wireless devices intended for use as sensor nodes in precision agriculture be allowed to share the 433 MHz sub-band currently authorized for use by active RFID devices at cargo terminals, port facilities and warehouses so that they may experience less path loss and achieve greater range and reliability while consuming less power.  In Chapter Four, we analyze 2450 MHz channel impulse responses that we measured in a high-density apple orchard and consider the implications of their form and scale for the design and deployment of WSANs and our understanding of the propagation environment. Of particular note is the vastly reduced delay spread compared to that observed in traditional residential, commercial and industrial environments. Finally, in Chapter Five, we summarize our contributions and offer recommendations for further work and next steps.   3Chapter 2: Deployment and Configuration of Wireless Sensor and Actuator Networks for Precision Agriculture  2.1 Introduction  In recent years, wireless sensor/actuator networks have emerged as a potential enabling technology for precision agriculture by facilitating sustainable techniques. Such networks help to tie sensors and actuators that are deployed throughout fields to the decision support systems that permit farmers to make effective choices to: 1) maximize production, 2) minimize waste/environmental impact, 3) mitigate pest infestations and disease, and 4) make the best use of scarce resources. Agricultural implementations of WSAN technologies are still very much in development, and important decisions regarding operating frequency, higher layer protocols, and deployment guidelines have not yet been made.  Most current deployments use unlicensed ISM frequency bands, however there have been no discussions of attractive licensed alternatives, and/or new allocations dedicated for agriculture. Commercially available nodes/devices use mostly 802.15.4 ? based standards (especially ZigBee, but also active RFID and Bluetooth). Current deployments typically use mesh/star network topologies but detailed investigations of optimal topologies, node spacing, etc. have not been conducted. Deployment guidelines regarding node placement, energy efficiency, and reliability have largely not been developed and would likely depend heavily upon the specific crop type and agricultural process/function.  2.1.1 Previous Surveys of PA and WSAN   Many surveys of advances in precision agriculture and the specific role of wireless sensor networks have been presented during the past twenty years. We summarize them here.  One of the first survey papers of precision agriculture (according to [1]) was presented by Schueller in 1992 in the journal of Fertilizer Research [2]. Schueller noted that many researches had recognized a need for ?spatially-variable control? of agricultural  4processes, as well as the need to understand and leverage the non-homogenous nature of crops, soil, and pests. At this time, automated agricultural sensors were just beginning to be developed. Also, remote sensing methods were beginning to be applied to agriculture, such as Synthetic Aperture Radar (SAR), which was used to measure topography, and infrared thermometry which was used to measure crop canopy temperature and water content.  Zhang et al. surveyed worldwide applications of the methods of precision agriculture (PA) in the journal of Computers and Electronics in Agriculture in 2002 [3].  In this survey, the authors discussed the concept of spatial and temporal variability in agricultural requirements and the use of PA mapping, sensing, and control technologies to respond to this variability. At the time, they noted that there was a lack of quantitative evidence of the positive impact of PA on profitability and environmental conditions. The authors briefly summarize many realizations of PA methods in regions as diverse as: Canada, Japan, Australia, Turkey, Saudi Arabia, Korea, France, United States, Argentina, Russia, Chile, The Netherlands, Italy, UK, Sri Lanka, Bangladesh, China, Brazil, Uruguay, and Indonesia. Zhang further notes that global adoption of PA had been slow and cites four barriers that must be overcome before a more widespread adoption could be realized:  a. Adopters were experiencing ?data overflow? which required the development of: data integration tools and decision support systems. b. There was a lack of rational procedures and strategies for determining application requirements on a localized basis and a parallel lack of scientifically validated evidence for the benefits claimed by the PA concept. c. PA methods required labor-intensive and costly data collection requiring the development of rapid sensing systems. Note that at this time, PA methods were implemented via manual data collection, remote sensing, or autonomous sensors/actuators. WSAN had not yet been widely applied to PA. d. There was a lack of technology-transfer channels and personnel. Educational programs involving researchers, industry, extension specialists, and consultants were also in short supply.  5 Cox presented a summary of then-current technologies used to implement PA in the journal of Computers and Electronics in Agriculture in 2002 [4]. He noted that remote and proximal sensing techniques (e.g., imaging, spectral reflectance, airborne laser based radar, and ground-based vertical-looking radar) had gained uses in measuring crop health, plant water content, soil status, weed/crop discrimination, and tracking insect and pest migration. Chemical sensors had also found applications in sensing soil nutrients, air conditions, and plant enzymes and antibodies. Cox also discussed the application of these sensing techniques and GPS devices to the monitoring and control of agricultural process.  McBratney et al. presented a survey of the most pressing issues for the adoption of PA and proposed several potential directions for future research. This survey was published in the journal of Precision Agriculture in 2005 [5]. The authors stated that the two most critical issues preventing the adoption of PA by agricultural producers were: a. Political dimensions: ?some see the technological focus of PA as a way of enhancing the hegemony of multinational farming corporations, thus some see dangers in its adoption in the developing world.? b. A lack of formal decision support systems to provide farmers with recommendations based on collected inputs. The authors also presented six other issues in PA that required more research: c. Appropriate criteria for economic assessment of PA. d. Insufficient recognition of temporal variation. e. Lack of whole-farm focus. f. Crop quality assessment methods. g. Product tracking and traceability. h. Environmental auditing.  Wang et al. published a survey of WSAN applications in agriculture in the journal of Computers and Electronics in Architecture in 2006 [6]. The authors summarized the requirements for wireless sensor technologies, namely: 1) robust radio technology, 2) low-cost, energy efficient processor, 3) flexible I/O for various sensors, 4) long-duration energy  6source, 5) small S/W footprint to run on a small processor, 6) S/W that makes efficient use of energy, 7) S/W with fine-grained concurrency, 8) S/W with high modularity, and 9) robust ad-hoc mesh networking. The authors then discussed the applicability of available wireless standards (ZigBee, WiFi, Bluetooth). They also presented a list of WSAN agricultural applications from the late 1990?s to the mid-2000?s, including: precision irrigation, spatial data collection (e.g., yield, soil conditions, temperature), variable-rate fertilization, and greenhouse control.  Ruiz-Garcia et al. published a survey of WSAN applications for agriculture in Sensors in 2009 [7]. The authors commented on advantages to wireless sensors over wired ones (e.g., cost, size, power, flexibility, distributed intelligence) and summarized modern architectures and standards (e.g., ZigBee, Bluetooth, RFID). The authors also presented a summary of deployments in a variety of PA sub-fields (e.g., pest control, irrigation, farm machinery control, and fertilization) in a variety of environments (e.g., viticulture/vineyards, greenhouses). Lastly, they acknowledged various aspects of the propagation environment that may affect link quality, including: seasonal changes in foliage size/extent, variations in plant density, weather/wind/moisture, and seasonal change in crop canopy height and density.  Gebbers and Adamchuck published a survey of PA in the February 2010 issue of Science [8]. Here, the authors summarized the application of soil-sampling, remote sensing (i.e., aerial or satellite imaging), proximal sensing (i.e., local electromagnetic, optical, and electrochemical sensors), and yield monitors to provide real-time inputs to decision-making processes. This survey did not mention the application of WSAN to PA.   Rheman et al. submitted a survey of agricultural WSAN applications in the journal of Computer Standards and Interfaces in 2011 [9]. Only a ?Corrected Proof? of this survey is available and it is unclear whether it was ever published. Here, the authors presented a detailed comparison of commercially available agricultural sensors such as: soil (temperature, moisture, dielectric permittivity, rain/water flow, water level, conductivity, salinity) plant/leaves (photosynthesis, moisture, hydrogen, CO2, temperature), weather (temperature, humidity, atmospheric pressure, wind speed, wind direction). They also  7discussed commercially available WSAN nodes and commonly used wireless standards. The authors further note several design considerations for WSAN, such as: energy consumption, sampling and data transmission strategies, fault tolerance, node size and weather proofing, and sensor placement, which has implications for reliability, energy consumption, validity of measured data, proximity to measurement subject. Lastly the authors list several aspects of modern agriculture that highlight the applicability of WSAN, such as: the need for real-time collection weather/crop/soil information, multiple crop-types grown in close proximity, and spatial variability in fertilizer and irrigation requirements.   2.1.2 Limitations of Previous Work  Previous surveys have covered: 1) available sensor technology, 2) existing WSAN products, 3) current and potential applications of PA WSAN, 4) potential benefits of PA over existing agricultural techniques, and 5) obstacles to widespread adoption of PA. However, some questions remain.  Propagation Environment: Of the surveys of PA and WSAN published to date, only Ruiz-Garcia et al. [7] have noted the particularities of the agricultural propagation environment that can affect the link quality of WSAN nodes. However, other researchers in the field have reported that: a. Energy consumption is a key design limitation for WSAN nodes [10] [11]. In fact, some researchers have found it to be ?the major problem in the operation of the [wireless sensor] network.? [12]  b. Between sensing, communication, and data processing, a sensor node expends the most energy through data communication [13] [14].  c. The major performance consideration in agriculture for low-powered wireless networks is large variability in signal quality, which ultimately determines packet reception rate. [15] d. For agricultural applications, WSAN must be able to operate in a range of environments, from bare fields to orchards, from flat to complex  8topography, and over a range of weather conditions, all of which affect radio performance. [15] Thus, understanding the propagation environment, and therefore the temporal and spatial variability in link quality is key to providing performance prediction, fair comparison of new technologies, and deployment strategies for PA WSAN applications  Demonstrations and Field Trials: Wang et al. [6] and Ruiz-Garcia et al. [7] have presented summaries of field deployments and real-world realisations of PA WSAN?s. However, it is not clear from these lists what trends have emerged over time (e.g., applications, environments, wireless technologies, sophistication, or statistical rigor). It is also not clear from these surveys whether there are any common outcomes from PA WSAN demonstrations and field trials to date, or what lessons have been learned that might benefit future researchers.  Network Performance Issues and Enabling Technologies: There are key differences between network performance challenges and requirements for agricultural and conventional WSANs. However, the majority of deployments and field trials conducted to date use commercial, off-the-shelf WSAN hardware and software. Many clever methods for improving PA WSAN performance have been proposed in the open literature that have not been covered in previous surveys. Perhaps the biggest open question is the potential for cross-layer design. Many other issues and challenges remain unresolved.  2.1.3 Objectives  In this chapter, we will conduct a survey of the open literature concerning WSAN as applied to PA with an emphasis on the air link and its impact on WSAN design. We seek to answer the following questions: a. What are the challenges posed by the unique propagation environment experienced by PA WSAN?  9b. What propagation models exist in the open literature which can be used for performance prediction, comparison of competing technologies, and simulation of PA WSAN? c. What trends have emerged from deployments and field trials of PA WSAN? d. What recent developments in the sensor network literature can be leveraged to increase performance and reliability of PA WSAN?  2.2 Emergence of Precision Agriculture  The techniques and methods of precision agriculture have been of interest to researchers and agricultural producers for the past forty years.  In this section, we present the history and current state of precision agriculture so that we may understand the current challenges/obstacles and how WSAN technologies offer an important solution  2.2.1 Current Challenges in Agriculture  Agricultural producers around the world are facing extreme challenges. Global population will continue to grow over the next 40 years and may peak at roughly 9 billion by the middle of the century [16]. Agricultural producers are experiencing greater competition for land, water, and energy, while also being pressured to curb the many negative effects of food production on the environment. Concurrently, high-quality farmland is becoming increasingly scarce and future agricultural expansion likely to occur on marginal, low-yield land which is vulnerable to degradation [17]. In many regions, soil quality is becoming increasingly degraded due to poor fertilizer and water management, soil erosion, and shortened fallow periods (in attempts to increase yield) [17]. In addition, many regions (e.g., China, India, Pakistan, Middle East, and North Africa) will soon fail to have sufficient irrigation water to maintain per capita food production [2] [18].   The inefficient use of fertilizers (e.g., nitrogen, phosphorous) has led to lower yields and increased pollution of surrounding soil, lakes, rivers, and streams [17] [19]. Also, the increase in global movement of crops, vehicles and people has led to increased proliferation  10 of non-native pests and diseases [20] [21]. These pests pose threats to biodiversity and ecosystem-level processes, causing losses in crops, forests, fisheries, and grazing capacity [22] [21]. Such infestations and outbreaks are the second leading cause of species endangerment and recent extinctions [21], and have an estimated cost to the US economy of approximately $137 billion / year [21].   2.2.2 Limitations of Traditional Solutions   The primary solution to food shortages has historically been to use more land for agriculture. However, over the last 5 decades, arable land has increased by only 9% [16]. This is largely because the expansion of agricultural land faces challenges due to: competition from other human activities (urbanization, biofuels), cost, protection of biodiversity and natural ecosystems, and losses of previously arable land (e.g., by desertification, salinization, soil erosion, and climate change). Efforts to increase yield by growing two or three crops per year on the same land have led to an increased proliferation of insects/pests/disease [17]. Other attempts to increase yield by increasing fertilizer use can lead to pollution and soil degradation, while failing to address field-to-field variation in soil-types and plant nutrition requirements [23]. Efforts to increase irrigation of agricultural land has led to inefficient use of worldwide irrigation water, and some global estimates state that only  37% of water delivered to the field is actually utilized by plants [18]. In many cases weeds, insects and other pests, and disease develop resistance to herbicides, insecticides and antibiotics (respectively) quickly due to overuse [17]. Many researchers (e.g., [23] , [3], [2], and [8]) have therefore noted the potential for ?Precision Agriculture? (also called ?Site-specific Crop Management?) to address these issues.  2.2.3 Precision Agriculture  By helping farmers make better, more focused, and more frequent decisions concerning application of remediation strategies, the technologies and techniques of PA are well positioned to address many of the current challenges faced by agricultural producers.   11 The National Research Council (NRC) defined PA in 1997 as: ?a management strategy that uses information technologies to bring data from multiple sources to bear on decisions associated with crop production? [24]. Others have defined it more broadly as: ?that kind of agriculture that increases the number of (correct) decisions per unit area of land per unit time with associated net benefits? [5].   Many researchers have proposed that PA is a promising solution to challenges described in Section 2.2.1. For example, in [23], Cassman et al. state that achieving consistent crop yields that exceed current limitations depends on sophisticated management of soil and water resources and applied inputs. Nutrient-use efficiency is therefore increased by better matching of temporal (i.e., correlated with real-time requirements) and spatial (i.e., at or near the plant roots) nutrient supply with plant demand and soil conditions. Such methods have the potential to reduce losses while maintaining or improving yields and quality [17] [23]. In [18], Qadir et al. state that irrigation water-use efficiency can be increased by delivering water directly to the roots frequently, in small amounts, at rates matched to plant requirements.   PA, also sometimes referred to as precision farming (PF) or site-specific crop management (SSCM), generally consists of a feedback loop comprised of the following stages [25]: a. Characterization: in which the extent, scale, and dynamics of variation are measured (i.e. sensing). b. Interpretation: in which the system assesses significance of measurements, identifies major causes of uncertainty, and formulates management targets (i.e., decisions). c. Management: in which inputs are applied at the appropriate scale and in a timely manner (i.e., action). d. Monitoring: in which the outcome of the actions are measured in a continuous learning process of change (i.e., observation).  12  PA applications have been motivated by different concerns in different agricultural regions [25]. In Europe, environmental issues and poor public perception of agriculture have driven PA applications to focus on more efficient use of inputs. In North America and Australia, pressure to increase the profitability of agricultural production and, to some degree, environmental issues such as nitrate leaching or phosphorus runoff have driven PA research. In some developing countries, simplified forms of SSCM have been created, driven by the need to produce more food, utilize inputs more efficiently, and increase farm profits in response to declining food prices.  2.2.4 Important Milestones in the Development of PA  The first step towards the techniques of precision agriculture was the development of ?geostatistics? in the 1960?s - 1970?s, which resulted from the study of small-scale (i.e., within-field) soil and crop variability [24] [26].   Then, in 1972, the launch of the first earth-observation satellite, LANDSAT, greatly advanced the field of remote sensing [27]. LANDSAT provided farmers with biomass, crop, and soil moisture sensing based on ?spectroscopic reflectance? (i.e., spectral analysis of the solar radiation reflected by plants and soils) [4]. This method was much cheaper and faster than the aircraft methods previously available [28] [29]. However, there were several limitations to earth-observation remote sensing, namely that use was dependent on weather (e.g., cloud cover) that a satellite pass only occurred once every 16 to 18 days [30] [31].  The first ?yield meters? were developed in the early 1980?s to record crop yield and analyze variability between fields [24]. Also in the 1980?s were the first uses of Variable Rate Technology (VRT) (i.e., fertilizer spreaders that could change the blend and rate of fertilizer on-the-go) [24]. The 1980?s also saw the advent of the first Geographic Information Systems (GIS) which were used to map spatial patterns in crop nutrients [24].   13 Global Position System (GPS) data was first available on tractors in 1991 with an accuracy of 100m and had improved to accuracies of 5-10m in the mid 1990?s [24]. Finally, in the early 2000?s, wireless sensor and actuator networks began to be deployed to assist in: spatial data collection, precision irrigation, variable rate technology (e.g., fertilizer), and supplying data to farmers (e.g., pest migration and disease infestation) [6]. WSAN provided numerous advantages over remote sensing and manual data collection, namely: 1) low cost, 2) ad-hoc, mesh networking (resilient to individual sensor faults), 3) automated data collection, and () ease of deployment via self-discovering nodes.  2.2.5 Current Challenges in PA   Early adopters of PA technologies (yield mapping, grid soil sampling, and on-the-go nitrogen testing) in 1980?s and 1990? suffered from [32]: a. A lack of education  b. Overly complicated, buggy software/UI c. Insufficient analysis of data: a lack of decision support systems left farmers unclear on what actions to take based on collected data d. Insufficient ROI: expensive technology during a less-than-profitable period in the agricultural industry Though the cost of technology has declined, many of these conditions persist and present barriers to adoption of PA methods by modern farmers.   Decision Support: Some researchers (e.g., [33] and [25]) have noted that the soil and crop science underlying many PA concepts may have lagged behind some of the rapid developments in agricultural equipment and information technologies. Researchers and farmers can now easily collect huge amounts of data, but assessing the quality of this information, transforming it into meaningful management decisions, and evaluating potential benefits and risks has proven to be a difficult task [25] [3] [5] [6] [34]. Others have noted that PA tools and techniques that have a clear focus on generating outcomes rather than information alone are more likely to be adopted by agricultural producers [35].   14  Efficacy: Surveys of farmers have found that one of the greatest disincentives to adoption of PA was a lack of evidence of increased yields, profits or environmental benefits [25] [35].   Cost: Researchers have also noted that the costs associated with upgrading existing IT infrastructure, purchasing new equipment, training and educating employees, and collecting and processing data present a significant barrier to the adoption of PA. [25] [3] [34] [35]  2.3 Wireless Propagation in Vegetated Environments: Literature Survey  The unique propagation environment in which PA WSAN are deployed will set the ultimate constraints on their performance and will affect: 1) coverage area/effective communication range, 2) optimal node spacing and network topology, and 3) required transmit power, and therefore energy consumption and battery life. There exists, therefore, a pressing need for propagation models which capture our intuition concerning the impairments that will degrade wireless system performance in a form that is useful in simulation and design.  We will summarize propagation models that exist in the open literature and assess whether their applicability to agricultural environments in detail in Section 3.3.4. Here, we seek to understand the unique characteristics of agricultural environments and systems, and the effects that these characteristics may have on performance and reliability.  2.3.1 Physical Nature of the Agricultural Environment  Agricultural environments are fundamentally distinct from the typical environments studied for sensor networks and PCS/common carrier systems (e.g., indoor, suburban macrocell). In this section we will describe the characteristics of these environments.  Time Varying: Unique to agricultural environment is the fact that the vegetation size, shape, and density can change significantly over the course of the growing season. For site-general models this will require estimation of model parameters (e.g., log-distance path loss  15 exponent) for each new environment, and the range of values that these parameters may experience throughout the season. For site-specific models (e.g., RET model in [36]) this variability requires a detailed description of the vegetation environment (type, density, and geometry/alignment)   Crop Types: Researchers have noted that the extent to which the presence of vegetation in the propagation path affects the amount of attenuation experienced by wireless signals depends greatly on the type and density of the vegetation (e.g., [36], [37], [38]). Thus, investigations of path loss for propagation through vegetation have been conducted in a variety of environments. Natural Forests are isotropic in nature, with randomly distributed trees, thick trunks, and stratified layers of vegetation. In [39] propagation measurements have been conducted in evergreen forests in which the trees are tall (25-30m) and spread out (0.03 trees per m2). In these environments, the thick trunks can cause severe shadowing of the signal. Further, observations suggest that if nodes are placed on the trunk, the coverage area in the NLOS direction will be reduced by ~ 50%.  Measurements have also been conducted in deciduous forests (e.g., [40] and [41]), in which trees are 5-10m tall, usually with a significant amount of underbrush (e.g., [41]). Artificial, or man-made configurations of plants tend to be anisotropic in nature. Propagation measurements have been conducted in a variety of orchards (e.g., plum [42], apple [15], and cashew nut [43]) in which the trees are shorter than in natural forests (e.g., plum ~ 2.7 m [42], cashew nut ~2.5m [43]) and are grown more closely together (1-3m apart), often in regular rows (e.g., [42]). Observations suggest that in these cases, transitions between the leaves-on and leaves-off state throughout the growing season may not greatly increase scattering and attenuation at wavelengths longer than average leaf dimensions [15], though increases in path loss (on the order of 5dB) due to the presence of leaves have been observed at 2.4 GHz [42]. Propagation measurements have also been conducted in plantations (e.g., palm, mango, and coconut [36]) in which trees are 5-8 m tall, and planted in a grid pattern (9m on a side). In these environments, the density of the canopy depends on the species (e.g., palm canopy has a much larger extent and higher density of leaves than mango canopy).  16 Observations suggest that the inhomogeneous vegetation density in these environments causes a large degree of spatial variation in signal strength [36]. Further, the degree to which a given model predicts path loss will depend on the propagation direction (e.g., in between rows, or along a row) [36].  Root vegetable fields (e.g., potato [44]) consist of plants which are grown in long soil ridges ~25 cm tall and 75 cm apart, and for which the above-surface portion of the plants reach heights of ~90 cm. Bushes and ground cover (e.g., rosemary, escallonia, creeping juniper [37]) grow low to the ground (~25-50 cm) and will vary in vegetation density depending on the species. For grain fields (e.g., wheat [45], corn [46]) there is a considerable change in vegetation extent and density over the growing cycle. Observations (e.g., [45]) suggest that radio range in these fields decreases significantly with crop growth (i.e., increasing vegetation density). Corn is typically grown in rows 60-80 cm apart, with about 15-20 cm between individual plants. Observations suggest that when corn is fully mature, the environment is fairly homogenous in vegetation density and propagation direction has little effect on path loss [46].  Plant Size and Antenna Height: Propagation characteristics may vary considerably with antenna height relative to vegetation. Studies of the effect of antenna height on path loss in agricultural environments can be broken into the following categories: a. Sub-surface (i.e., through-soil) propagation: Here, researchers have investigated sub-surface propagation for soil sensing applications (e.g., when both nodes are below the surface [47], when one node is below the surface and one is above [48]). For propagation through soil, path loss increases with increasing volumetric water content [48] [47]. Path loss was seen to increase by 10-20 dB between 433 MHz and 868/915 MHz [48]. At 433 MHz, the effective communication range when both nodes were placed in the ground was less than 70cm [47]. Therefore, for the case when sensors must be placed below the surface of a field, the antenna should be mounted away from the sensor, above ground  17 b. At ground level: Observations suggest that placing sensor nodes at ground level drastically reduces effective communication range (e.g., [45]). In potato fields coverage area for antennas mounted ground level was found to be ~23 m for low crop density (early and late in the season) and 10 m during peak density (flowering stage) [44]. For low shrubs and ground cover, maximum node separation at ground level was found to be only a few meters (i.e., < 10) [37]. In wheat fields (during the seeding stage), radio range for ground-level nodes was observed to be less than 10m [45]. In corn fields with regular rows, effective transmission range was roughly 20m for sensor nodes at ground level (2.4 GHz) [46]. Therefore, If nodes must be separated by more than 10m, the antenna must be mounted higher than the surface of the soil c. Below canopy: Researchers in [39] observed a 10-15m increase in range when the TX node was increased from 1.5m (height of RX node) to 3.5m (based on SNR/BER of 802.15.4) when using commercial WSAN hardware (at 2.4 GHz) in pine forests (i.e., both nodes were well under the canopy layer). However, for researchers in [38], antenna height (3-5m) was found to have little to no effect on path loss exponent for 1.8 GHz signals propagating in forested environments (tree heights between 7-17m). Comparison of commercial WSAN motes at antenna heights of 0.15m and 1 m in cashew nut trees (average height of 2.5m) revealed that signals experienced twice the path loss when the antenna was close to the ground (i.e., the log-distance path loss exponent was twice as large at 0.15m than at 1m) [43]. Measurements taken in wheat fields at 2.4 GHz revealed that radio range is less than 50 m for antenna heights lower than 0.5m [45]. For 2.4 GHz signals propagating through cornfields (max crop height ~3.2m), signal strength varied greatly (10-20 dB) with antenna height [46]. Optimal antenna height was observed to be roughly half the crop height (1.6m). The authors suspect this is because both ground reflections and canopy reflections are minimized.  18 d. Above canopy: Measurements taken in wheat fields at 2.4 GHz suggest that radio range generally increased with antenna height (once above the canopy) at any point in the growing season (though increasing from 1.5m to 2m saw a small decrease in range) [45]  Plant Structures and Frequency Effects: As the researchers in [42] and [15] discussed, absorption and scattering of wireless signals by vegetation is likely to increase for wavelengths shorter than leaf and branch dimensions. Therefore, greater spatial variability in received signal strength will be observed at short wavelengths (relative to local scatterers) (e.g., [49]). Researchers in [50] found that the excess attenuation due to vegetation was significantly higher at 11.6 GHz than at 1.2 and 2 GHz in the leaves-on state. However, the differences in excess attenuation between high and low frequency were much smaller in the leaves-off state. In addition to attenuation effects, as wavelength increases the height of the first Fresnel Zone will increase, leading to stronger ground reflections and therefore increased spatial variation in signal strength [36]. For antennas mounted above the crop canopy, a shallower first Fresnel Zone will decrease the antenna height required to minimize clutter losses, however free space path loss also increases with frequency.  Crop Geometry and Propagation Path: Researchers in [46] compared propagation direction in corn rows (at angles of 0?, 45?, and 90? relative to rows) for 2.4GHz signals with antennas mounted at roughly half the max crop height (1.6m). They found that propagation direction had little-to-no effect on path loss, likely because of the homogeneity of the environment (at full maturity, plant growth fully obstructs rows). For inhomogeneous environments, the amount of path loss experienced may depend on propagation direction. For example, in [51] less path loss was observed for propagation in between rows of palm trees than for propagation along rows.   Climate: A number of researchers have observed increased attenuation due to moisture on the leaves of vegetation in the propagation path. In [40], measurements in a forest of Sycamore, Ash, and Oak (all deciduous) forests at 38 GHz indicated that the presence of moisture increased the attenuation due to vegetation by roughly 4 dB. In [41], an additional  19 loss of 7dB was observed as a result of moisture on the leaves of a single chestnut tree at 13 GHz. In [52], a rain rate of 6mm/hr increased the measured path loss through a grove of poplar trees at 2.5 GHz by about 5 dB. The amount of attenuation due to moisture is therefore dependent on frequency and density of vegetation   2.3.2 Characterization and Modeling in Other Fields of Study  Agricultural environments have been studied by researchers in other fields of study. For example, agronomic researchers are principally concerned with describing/characterizing the tissues, organs, water content, and economic yield of crop plants ( E.g., [53], [54]). However, some (E.g., [55]) have developed statistical models of plant structure orientation and distribution. Researchers in remote sensing (E.g., [56]) use VRT to statistically describe vegetation and thereby model backscatter from crop canopy layers (high incidence angle from satellites, planes). The authors in [57] have discussed how these models change throughout the growing cycle (in this case, for wheat and sunflower plants). An attractive possibility exists, therefore, to leverage the statistical models of plant structures developed by agronomists and combine them with the techniques of VRT to accurately describe propagation through vegetation and thereby enable those tasked with developing/deploying WSAN for agriculture.  2.4 Demonstrations and Field Trials of WSAN in PA Applications  In this section we present a representative sample of the real-world deployments of Wireless Sensor and Actuator Networks (WSAN) in Precision Agriculture (PA) applications that have been published in the open literature to date. Thus far, the large communications players have not entered this space. Demonstrations and field trials have been performed by university researchers, government labs, and agricultural corporations.   In this section we have surveyed both the wireless and agricultural literature to obtain a representative sample of the academic works published to date. Our goal is to identify: 1) the agricultural applications for which WSAN have been considered/studied, 2) characteristics of  20 the sorts of environments PA WSAN must operate in, and 3) the wireless technologies that have been leveraged to deploy PA WSAN. We will also describe the evolution of the work in this field over time, including: 1) research project sophistication, 2) milestones, and 3) key outcomes and lessons learned.  2.4.1 Role of Demonstrations and Field Trials  The published demonstrations and field trials presented here have been conducted by researchers and developers for a variety of reasons:  Familiarization: In this case, researchers are concerned primarily in familiarizing themselves with available technology and the relevant design challenges, while also publishing a  statement of intent (e.g., [58] [59] [60] [61] [62]). These papers are often published as a precursor to or in parallel with prototype development. In this phase, researchers: (1) consider system architecture and software/hardware design (e.g., [58]), (2) develop routing and control algorithms (e.g., [60]), and/or (3) perform simulations of the proposed systems (e.g, [59]).  Proof-of-concept / Prototype development: Here, researchers actually develop and deploy one or more nodes to test the proposed design (e.g., [63], [64], and [65]). In these studies, node operation may be tested individually on bench-tops (e.g., [63]), in networks in idealized or sterile environments (e.g., [11]), or in real-world scenarios (e.g., [66] and [67]). In this phase, researchers are primarily concerned with validation of design decisions, investigating energy consumption and system performance, and optimizing hardware and software architecture.  Cost-benefit analysis / Feasibility study: Here, researchers assess the applicability of WSAN?s to PA applications. These studies investigate: the reliability of node communication (e.g., [68] [69]),  whether fine-grained data collected by WSAN?s can provide better inputs to PA DSS than conventional methods (e.g., [70]), the validity of collected data (e.g., [64]), and whether WSAN are cost-effective (e.g., [71]).  21  Sampling strategy / Data acquisition analysis: Here, researchers are concerned specifically with temporal sampling rates and the implications of node-spacing and network topology on spatial sampling. (e.g., [68], [72], and [73]).  Network design and development: Here, researchers are concerned specifically with the effect of network topology, node spacing, and foliage growth on error rates, node drop-outs, and link reliability (E.g., [67], [74], and [48]).  Commercial deployments: There are a few cases in the open literature where the systems described are or have been used in commercial applications (e.g., [12], [75], and [76]).   2.4.2 Agricultural Applications  Wireless sensor networks have been deployed in a variety of agricultural environments to facilitate a variety of processes using a number of different wireless technologies. In this section we seek to quantify and categorize these publications to obtain an understanding of the development of this field to date. In Figure 1, we can see that the majority (~30%) of these published deployments have not been targeted at any specific agricultural process. The most common process to which these deployments have been applied is precision irrigation, comprising about 28% of the publications.   Figure 1 ? Proportion of demonstrations and field trials which have been conducted for each agricultural Pest/disease detection/treatmentto collect real-time micro-climate and soil data that can, in turn, be used to provide inputs to PA DSS that assess the risk of pest infestation then target problem areas with pesticides/fungicides/herbicidesin conjunction with pest traps to relay real Precision irrigation (monitor and control) (E.g., [65] [48] [83]): Precision irrigation increases crop yield and reduces runoff and leaching by precisely matching water supply to the spatially distributed crop water demands caused by variability in soil properties and topognecessary real-time inputs for such  VRT control (e.g., fertilizer) measurement data from soil sensors can be collected by control of wireless actuators which regulate fertilizer and fertigation systems.  Frost/freeze detection [12] [70low-temperature injury to plants. Growers of hightherefore required to use various methods of active protection (heating, irrigation, air mixing) application   (E.g., [68], [69], [75], [77]and [60]): WSAN[77] or the outbreak of disease . WSAN have-time infestation alerts to farmers [[78] [75] [79] [80] [59] [11raphy [78]. WSAN have been used to site-specific control and management systems [84] [76] [61]: Similar to the precision irrigation problemWSAN and used as inputs for ]: In some agricultural regions climate poses a severe risk of -value crops (wine grapes and fruit) are 22   have been used [68] [69] and  also been used 77] ] [81] [82] [61] provide the [78]. , [61]  23 to prevent frost or freeze damage to their fruit [12]. Conventional methods require workers to continually monitor air temperatures, initiate and operate protection systems when critical low-temperatures are reached, and shut off the protection systems when regular growing conditions are restored. This is very expensive in labour and operating costs [12]. Researchers have found WSAN to provide cost-effective and reliable real-time inputs to low-temperature crop protection DSS [12] [70].  Fire Detection: Early detection and suppression of plant fires is crucial for restricting the amount of damage they cause [66]. WSAN can alert fire-fighting agencies to the presence of forest fires and plant fires before they are able to spread to other regions [66].  Storage monitoring: Researchers have used WSAN as a non-invasive method of monitoring silage for livestock feed [85] [86]. WSAN could be a useful alternative to more invasive, traditional methods which can have a negative impact on silage quality by bringing the silage into contact with O2 and thereby causing decomposition of the digestible matter [86]. Demonstrations have shown that such wireless sensor nodes can successfully be used to detect silage decomposition and therefore improve the efficacy of silage conservation systems [86].    Maturation Detection: Researchers using WSAN in viticulture applications have demonstrated that ?heat unit accumulation? data (temperature over a baseline summed over some period) can be used to predict standard measures of fruit maturity [70]. Such systems could therefore be used by grape growers for sampling strategies, picking decisions, or fruit segregation for fermenters. [70]  Livestock tracking [74] [87] [72] [88]: WSAN can provide cost-effective means to monitor livestock grazing habits and drive decisions to increase land-use efficiency [74]. Recent outbreaks of disease among livestock populations (e.g., Bovine Spongiform Encephalopathy and Foot and Mouth Disease) illustrate the need for timely animal health monitoring [87]. Wearable node collars can allow farmers to track livestock and monitor health with greater frequency and regularity than would be practical with farm staff [74] [87]. The body of the  24 animal can cause shadowing and severely attenuate wireless signals. Researchers have used spatial diversity antennas to help mitigate this drawback [74]. In the case of intraruminal sensors, which allow livestock producers to monitor temperature, pressure, and pH level of the digestive systems of individual cattle, the wireless nodes are subject to a significant amount of translation and rotation. Researchers have investigated the feasibility of using relay nodes atop the animal to route data from the internal sensors to the base station [88]. Because animals are mobile, the network topology is constantly changing, therefore routing algorithms may require more frequent discovery phases [74]. The physical extent of farms can also provide a problem for coverage. Typical farm sizes in the UK are in the tens of square kilometers, and hundreds of square kilometers in Australia and North America. Therefore, livestock monitoring WSAN will likely require several stationary repeater nodes to guarantee reliable coverage. [87]    Microclimate sensing (non-specific app.) [63] [58] [67] [89] [90] [91] [64] [92] [71] [73] [93] [94] [62]: A number of published deployments have merely collected climate data with no focus on a specific agricultural application.   2.4.3 Environments  In Figure 2, we see that the most common environments in which agricultural been deployed are vineyards (17%), fruit orchards (14%), and greenhouses (12%). These environments produce high value crops and are therefore likely to provide a higher return on investment for WSAN deployment. Figure 2 ? Potato fields: Researchers have deployed treat outbreaks of phytophtora fungus humidity, temperature, and the amount of moisture present on the plants? leaves sensing cannot be used to obtain the potato crop [44]. Potato plants are grown in ridges of earth in parallel rows and reach a maximum height of about 90 cm during the growing seas[44].One group of researchers found that radio range for reliable communication (at 433 MHz) when using Mica2Dot nodes flowering [44] .  Fruit Orchards/Fields: Fruit orchards present unique impairments to wireless signals as a result of the physical extent of the constituent trees.  Proportion of deployments in each environment WSAN in potato fields in an attempt to detect and [68] [69]. Key factors in the growth of this fungus are information about the microclimate beneath the canopy of on when flowing in midwas only about 10 meters when the potato crop is  They also present the possibility of a 25 WSAN have  [68]. Remote -July  26 high return on investment as they typically produce high-value crops. Researchers have conducted test WSAN deployments in apple (e.g., [79] and [76]), kiwi (e.g., [76]), eggplant (e.g., [66]) and other (un-identified) (e.g., [12] [90]) orchards/fields. Common risks to orchard crops consist of extreme temperatures [12], pests [76], and fire [66].  Vineyards: WSAN?s described in [12] and [95] have been used to monitor soil moisture, conditions for pests and disease, crop load in grapes, and frost/freeze event detection in vineyards. WSAN?s in [75] have been deployed at 4 pilot vineyards in Europe and have been monitoring soil, climate, and plant conditions since 2005. Vineyards in Spain presented unique challenges for deployed WSAN?s due to the hilly nature of the region. There, researchers were able to monitor crop, soil and climate conditions using the 868 MHz ISM band using repeater nodes and directional antennas to overcome the complex topography [84].  Use of WSAN?s in vineyards in [70] revealed that: (1) heat summation data provides growers with better awareness of fruit maturity than standard methods (elevation heuristic), (2) a less dense network (15m for leaf nodes, 25m for backbone  nodes) would have been adequate to accurately predict grape maturity, (3) temperature data provided farmers with sufficient warning to prevent frost damage. A WSAN deployed in an Italian vineyard also allowed wine-producers to take corrective action when a fungal infection was detected [73].  Olive Groves: A network of nodes utilizing a trap and camera sensors was used to detect pests in an olive grove in Croatia. Researchers were able to correlate the pest infestation with microclimate data (also collected by the WSAN) [77].  Greenhouses: A number of researchers have tested prototypes of WSAN?s deployed in greenhouses which provide inputs to DSS for application of irrigation, fertilizer, and lime and pesticides [70] [71] [60][93][62].  Soybean fields: A prototype system was deployed in 800 square meters of soybean fields in China in 2011 [92].  27  Cattle farm: WSAN?s have been deployed for monitoring health, location, and grazing patterns of livestock. Researchers in the UK, in [74] and[87], found that: (1) the use of a multi-hop routing protocol was sufficient to mitigate shadowing caused by animals in motion, and (2) that a spatial diversity scheme (antennas either side of head) provided greater coverage reliability.  Researchers in Denmark [72] found that they were able to draw inference to the position of the whole herd by monitoring a subset with sensor nodes. Researchers in Australia [88] were able to collect data from cattle via internal sensors and transmit to external routing nodes and thereby monitor health conditions.   Cotton Fields: Researchers in [65] deployed a WSAN in a cotton field to measure soil moisture content for scheduling irrigation.  Cabbage Fields: In [94], nodes deployed at ground level in a cabbage field lost communication as density of foliage increased throughout the growing season. The researchers found that using antennas with higher gain allowed the nodes to communicate at distances of 50m throughout the season.  Grain Silage: Researchers in [86] used WSAN to monitor grain silage decompositions without compromising the airtight seal of the stacks, which is common with traditional invasive monitoring systems.     2.4.4 Wireless Technologies  The distribution of wireless technologies used to connect PA WSAN nodes in the deployments summarized in the preceding sections is shown in Figure 3. By far, the most common communications protocol used is ZigBee, which accounts for about 45% of the published deployments to date. Other 802.15 networks with different NET and APP layers, or with MAC layer modifications (compared to the ZigBee implementation) make up about 20% of the published literature. Bluetooth and Active RFID (802.15.4f) each were only used  for a single deployment while proprietary protocols and TinyOS (typically used by UC Berkley and Crossbow MICAreported here.  Only about a quarter of the demonstrations and field trials above mentioned the use of a backhaul technology to connect the networks to a centralized control, processing, or storage facility (see Figure 4). Of those, GSM andWiFi accounting for the remaining 20%. Figure Figure  2.4.5 Growth and Sophistication In this section we seek to describe the growth of the field of PA sophistication of each deployment described above and any analysis presented by the authors. As can be seen in both -based nodes) each comprised about 10% of the deployments  GPRS were each used about 40% of the time, with  3 ? Wireless technologies used by nodes 4 ? Wireless technologies used by backhaul  WSAN by categorizing the Figure 6 and Figure 8 deployments of this sort began being 28    published in the open literature around 2004 and have reached a peak in recent years. The low number of deployments in 2012/2013 is likely a result ofcoupled with the time this survey was conducted (late 2012/early 2013) and should not be interpreted as a lack of interest in the field. Realization: Categorization of the level of sophistication of each field tshown in Figure 5 and Figure conducted in real-world agricultural environments. About half of those studies were conducted over a period of time less than a growing season, while the other half were conducted over the entire length of the season. About 20% of the studies involved only unit level testing in laboratory settings and about 10% contained no deployment whatsoever.Figure Figure 6 ? Sophistication of trial   the inherent publication delay  rial describ6. Here we see that the majority (~60%) of studies were 5 ? Sophistication of trial WSAN realizations WSAN realizations over time 29 ed here is     Analysis: Categorization of the degree of sophistication in the analysisshown in Figure 7 and Figure provide no analysis of the performance or viability of the deployed network beyond anecdotal assessment of whether thedeployments mentioned some measure of observed node failures, error rates, or communication range. A final 17% made recommendations of deployment guidelines, or models, or assessed the applicability of Figure Figure   of each field trial is 8. Here we see that the majority of published studies (~49%)  network was able to collect data. About 34% of WSAN to the PA process under study.7 ? Sophistication of WSAN analysis 8 ? Sophistication of WSAN analysis over time 30     31 2.4.6 Lessons Learned  The authors of the studies considered here have made a number of observations that are useful to other developers in the field. They found, for example, that: a. Unique S/W and H/W design challenges can result in node failures, and poor data delivery performance [68][69][70][83]. b. Power consumption is a crucial limiting factor in node/network design [12][95][78][89]. c. Data transmission via wireless communication accounts for the largest portion of nodes? energy consumption [63]. d. Communication range of the nodes can vary greatly depending on: HW design [63][12][91][94], antenna selection/design [82], frequency band [48], and crop growth [66][65][94]. e. Topology can be a challenge for deployment strategy [84]. f. Data storage can be an issue for long term deployment [76]. g. The spatial and temporal variability of data should be considered when designing node deployment and sampling strategy [70][72]. h. WSANs can still be cost-prohibitive in some applications, with some nodes ranging from $100  - $1000 [64][71][65] i. WSANs can provide great improvement over wired (or other traditional) methods [70] j. WSANs are already being used successfully by farmers to meet agricultural challenges [12][75][76][70][73]  32 2.5 Performance Issues and Enabling Technologies  In this section, we survey the wireless sensor network literature and report on the potential of recent developments and advances to increase the performance, efficiency, and reliability of agricultural networks. Common metrics of WSAN performance include energy efficiency, end-to-end delay, and throughput.   In any sensor network wherein nodes operate away from convenient sources of power, energy consumption is a significant concern, and some have found that as much as 80% of the energy used by nodes is spent on data transmission [96]. In addition to energy consumption, reliability and connectivity are the primary design requirement for networks in precision agriculture. The key challenge for reliability that is specific to PA implementations of WSAN is that nodes must be installed in harsh environmental conditions and must function throughout the length of the growing season. The density and extent of vegetation may also change considerably over the length of the season, which presents a unique challenge for nodes to maintain connectivity throughout the year [45].  Many researchers have noted the need for reliable communication and long lifetime have nodes (e.g., [93], [82], and [94]), however there has been little study of expected battery life or packet error rates experienced in agricultural environments. In [93], Zhang et al. observed that their nodes lasted for 6 months on 2 D-batteries however no analysis of the power consumption or factors that contribute to it was given. In [82], Panchard et al. report a lifetime of 1-2months achieved with proprietary crossbow MICA2 motes provide no details as to how the proprietary MAC layer used contributed to battery life. In [94] Lopez et al. describe a deployment of CC2420 chips using IEEE 802.15.4 networks which resulted in a lifetime of 10 weeks. A study of the impact of routing on network connectivity for cattle management is presented in [87].       33 2.5.1 Node Placement and Network Topology  The simplest network deployment strategy employs a regular grid spacing, and it is this topology that has been used for the majority of agricultural WSAN deployments to date [97]. However, a few researchers have described other deployment strategies and have discussed the effect of gateway node placement (e.g., [98]) and the use of other topologies in decreasing the total number of sensor nodes used to cover a given region (e.g., [99]).   In [45], the authors compare equilateral triangle, square grid, and regular hexagonal spacing in terms of cost, and connectivity. They found that triangular spacing was the most efficient in terms of coverage area (i.e., had the lowest cost) but also had the lowest connectivity. Regular hexagonal spacing was found to have both the highest cost and the highest degree of connectivity. In [99], more sophisticated topologies were proposed based on the variability of various agronomic parameters (e.g., salinity, water content and nutrition) [11].  2.5.2 Data Compression and Node Management  In [98], the authors have proposed an in-network reduction technique that increases network lifetime based by taking advantage of temporal correlation of the sampled data, (e.g., microclimate measurements). The authors state that this technique is far less complex and memory intensive compared to other approaches, and can be easily implemented on MICA brand motes.  In [100], successive approximation multistage vector quantization (SAMVQ) was proposed in order to predict the chlorophyll content of crops for precision agriculture. The field under study was assessed in nine different zones based on the amount of fertilizer needed for each zone. The authors claim the technique to be simple and fast, with small loss and a high compression ratio.  In [13], the authors propose a decision rule for keeping nodes active or shutting them  34 off based on predictions made by neighboring nodes. In the proposed method, special parent nodes make these decisions for children nodes by applying linear regression to their data in an effort to predict future samples. The parent nodes consume slightly more energy due the added processing, but reduce overall energy use by keep the children nodes in ?sleep? mode. This prediction mode can be turned on depending on the accuracy of predictions and remaining energy stores of nodes [13].     In [10], Mainwaring et al. note the importance of data compression for their sample PA WSAN deployment. They estimate that the volume of data can be reduced by a factor of 2-4 by applying standard data compression algorithms (e.g., Huffman coding or Lempel-Ziv).    2.5.3 MAC Layer Techniques  In [101], Ferentinos et al. propose a Genetic Algorithm (GA) based MAC protocol. In the proposed protocol, the GA decides which nodes in the field should stay active or should be assigned as cluster heads based on the fixed placement of nodes. A fitness function is derived based on the PA application-specific parameters of interest (e.g., mean relative deviation (the measure of samples? uniformity), connectivity, or battery life and operational energy consumption). However, the proposed method is static and does not adapt to changes in the conditions in the network over time.  In [102], Sahota et al. propose the Ping-Drowsy MAC (PD-MAC) protocol, which has multiple levels of transmit power, and sensitivity which can be selected from. The protocol allows nodes to receive messages at a low level but for a longer duration in order to make up for the clock drift which may happen due to long sleeping intervals.   In [69], Langendoen et al. have used the T-MAC protocol because of its adaptive duty cycle and ability to conserve energy. Moreover, early sleeping causes T-MAC to be more energy efficient compared to S-MAC at the cost of lower throughput.    35 In [103], El-Hoiydi et al. propose the WiseMAC (Wireless Sensor MAC) protocol which aims to lower power consumption for low data rate applications. WiseMAC updates access points with the scheduling offset of all sensors.  Significant improvement is observed compared to the IEEE802.15.4 MAC protocol in terms of energy efficiency, however, WiseMAC suffers from a lack of scalability and an absence of synchronization.    In [104], Chiti et al. propose an energy efficient cross-layer MAC protocol, called synchronous transmission synchronous reception (STAR), for outdoor environment monitoring. STAR MAC is an extension to WiseMAC and aims to reduce the energy consumption of WiseMAC by introducing a duty cycle. This scheme is a combination of S-MAC and WiseMAC and has been tested in the Agro-Sense project [61] in vineyards.  In most of the demonstrations and field trials in Section 2.4 networks use the IEEE802.15.4 MAC protocol (specifically ZigBee). Some use proprietary MAC protocols, such as in [75] where Texas Instruments? CC1000 radios are used. In [68] and[105] T-MAC is used which is more energy efficient than IEEE802.15.4.  In the [10] effect of S-MAC and ALOHA is compared. The deployment in [79] uses TinyNode584 technology which is a proprietary module made by shockfish/EPFL supporting data rates of 1.2-152.3kbps. In [91], and [83] nRF905 modules developed by Nordic are used, which allow use of TDMA, CDMA, and FDMA techniques in the 433, 868, and 915MHz bands. In [81] UFM-M11 proprietary RF modules are used, and no description of the MAC protocol is given.   2.5.4 Network Layer  In precision agriculture there are applications with both periodic (e.g., real time monitoring of crops/fields) and event-driven traffic models. Event-driven networks must have very low delay, while periodic systems could tolerate longer network latency. Networks in agricultural applications could be small (tens of nodes) if individual fields are being monitored, or very large (thousands of nodes) for whole farm coverage.    Mobile relay nodes may help to lower power consumption and increase network  36 lifetime as proposed by El-Moukaddem et al. in [106]. They show that as data size increases, both topology changes and mobile relays will be beneficial, whereas static topology and relays are more suitable for small size data chunks.  In [102], Sahota et al. proposed a networking layer technique based on dividing the  field into smaller subfields and placing base stations on the subfields corners (i.e., redundancy of base stations). With this approach, nodes can switch among base stations for every round of the data transfer which results in a balanced energy depletion of nodes.   In [69], Langendoen et al. use MinRoute, which is developed by UC Berkeley, as it has lower memory requirements and reduced complexity over  AODV. MinRoute configures a spanning tree towards the sink node.  In [107], Karthickraja et al. propose a rapid spanning tree protocol which aims to increase network configuration speed by switching between cluster heads and changing network topology which results in a more balanced depletion of batteries.    In [61], Anurag et al. propose a new addressing algorithm which aims to prevent the waste of address (as in ZigBee) and maintain an efficient routing procedure. The authors address the deficiencies of the ZigBee Cskip algorithm for address allocation and propose an idea based on the precision agriculture assumption that a hierarchical scheme is preferred for sensing and tracking applications since the path of sensing or tracking is known and data packets always flow between end device and coordinator [61].    In [108], Aquino-Santos et al. propose a location-based routing algorithm with cluster-based flooding which has modified to meet the requirements of precision agriculture. Temperature and humidity sensors in a small-scale network were deployed for this study. The routing strategy is explained for both flat and hierarchical structure. The routing algorithm is evaluated based on route discovery time, packet delivery ratio, end-to-end delay, throughput, routing load, and overhead. The hierarchical structure proves to be more scalable, whereas the flat algorithm is superior with regards to route discovery time, End-to- 37 End delay, routing load, and overhead. Therefore, for small-scale networks, the flat algorithm is preferred.   In [109], Sutar et al. propose a routing method which they call ?Ant Aggregation?.  In this algorithm, there are two variables called the pheromone trail and the pheromone model. Pheromone trails are stimulated with the help of models, which is a set of parameter values. The algorithm is inspired by the manner in which ants attempt to find the shortest path to sources of food.   In the majority of the demonstrations and field trials in Section 2.4, mesh or STAR network topologies have been used. However, in [59], Feng et al. used LEACH, which is a cluster-based routing protocol, and in [74] Kwong et al. use a multi-hop routing protocol for livestock monitoring.   2.5.5 Potential for Cross-Layer Design  Even though several network communications techniques have been proposed in the literature, there are still many open questions for future work. In [110] and [111] the authors use well-known propagation models (e.g., free-space, two-ray, shadowing, and small scale fading) to assess the impact of channel impairments on popular ad-hoc routing protocols (e.g., AODV, DSR, and OLSR). They observed that network throughput, end-to-end delay, and energy consumption are degraded significantly in the presence of channel impairments. The impact of the propagation environment faced by agricultural WSAN?s on higher layer technologies has not yet been investigated. Therefore, there exists a considerable opportunity to optimize higher-layer protocol design for the unique propagation challenges faced by these networks.  However, as noted in [69], no MAC layer protocol likely to be suitable to all applications. Moreover, designers choose the desired MAC protocol depending on application requirements, network topology, radio characteristics, and traffic model. Many MAC protocols have been proposed for low data rate applications in WSANs among which  38 S-MAC, T-MAC, B-MAC, LMAC, X-MAC, WiseMAC, and Crankshaft are the most notable ones.  Even though models for the latency and power consumption of these protocols are provided in[69], and elsewhere, the performance of these protocols under realistic precision agriculture scenarios (e.g., different seasons and climate conditions) is yet to be investigated.    39 Chapter 3: Spectrum Allocation for Short Range Wireless Devices in Precision Agriculture  3.1 Introduction 3.1.1 Significance of Spectrum Selection  Commercial WSAN technologies that are currently available mostly operate in the 2.4 GHz globally harmonized ISM band that offers broad bandwidth (100 MHz) and license-exempt operations on a non-protected, non-interference basis. Some also use the 902-928 MHz band (ISM in Region 2) or the 868 MHz band (ISM in Region 1). License-exempt bands have been beneficial for research and proof-of-concept deployments; however, it is worth considering whether alternative frequencies might be more attractive for large-scale commercial deployments.   Similar to other wireless services where devices must operate away from sources of power, agricultural networks must maximize coverage area while minimizing the energy consumption of nodes. The coverage area of a node will, for fixed transmit power, be determined by: 1) the effective area of an antenna (i.e., the frequency dependent component of free-space path loss), and 2) the degree to which obstacles in the propagation path (e.g., branches, stalks, leaves) absorb and/or scatter the wireless signal. Both of these mechanisms are proportional (inversely and directly, respectively) to carrier-frequency.  Bands of spectrum allocated to a particular service in one ITU-R region may not be available for use by the same service in a different region. Therefore, the selection of carrier frequency also determines the degree to which devices can operate in multiple regions. Choosing a band that is globally or regionally harmonized encourages manufacturers to jump on board, facilitating seamless global adoption of these wireless technologies. It is critical to consider these issues now, so regulatory action can begin, leading to global harmonization of frequency bands   40 3.1.2 Trends in Wireless  Bandwidth vs. Power Consumption. Over the last twenty years, development of wireless services has been driven by one of two requirements: 1) maximization of capacity through increased bandwidth available at higher frequencies (e.g., IEEE 802.11 in 2.4 GHz ISM and 5 GHz U-NII bands), or 2) minimization of energy consumption by limiting bandwidth and transmit power (e.g. IEEE 802.15.4). In the case of agricultural WSAN, achieving transmission over long-distances by narrow signal bandwidth is more important than high-speed transmission with broad signal bandwidth [112]. This is fundamentally different from the case of common carrier and personal communication systems where obtaining sufficient throughput/capacity is of primary concern.  High Frequency vs. Low Frequency. Trend towards using lower frequencies to leverage superior propagation characteristics (e.g., TVWS frequencies[113], 700 MHz LTE) especially for low power devices (e.g., ZigBee at 915 MHz and 2.4 GHz, and Active RFID at 433 MHz)  Spectrum Sharing. Regulators are increasingly promoting the shared use of spectrum, in part due to: 1), the recent exponential growth in wireless data traffic, 2) the successful use of license exempt bands (e.g., WiFi), and 3) the development of dynamic spectrum sharing technologies (e.g., cognitive radio). [114]  Harmonization. As already noted, regional and global harmonization of spectrum allocations allows manufacturers to reach economies of scale which, in turn, drives technology adoption. Harmonization also allows devices to be used in multiple regions, which is especially valuable for roaming applications, and in border areas. As a result, regulators are increasingly interested in allocation efforts that result in harmonization [115].    413.1.3 Objectives   In this chapter, we seek to: 1) identify attractive frequency bands for agricultural sensor and actuator networks, 2) review the requirements and typical use-cases for these applications, 3) survey the technical literature for studies of wireless propagation in agricultural environments, and 4) thereby determine which spectrum allocations are most appropriate for these unique constraints and environments.  3.2 Requirements and Challenges  3.2.1 Environment  The agricultural environment differs considerably from other industrial environments in which low power and low-throughput wireless networks are typically deployed: 1) Fields, crops, and orchards are cluttered and obstructed environments that attenuate wireless signals considerably, 2) Agricultural environments are extremely diverse and may be composed of low bushes and shrubs, root vegetables planted in raised soil ridges, regular rows of stalks, uniform fields of wheat or other grains, dense orchards, or forested plantations of tall thick trunks with dense, leafy canopies; in many cases the vegetation density can change considerably throughout the growing season, and 3) distinct from warehouses, factory complexes, distribution facilities, and container yards, farms are typically operated by a single user and deployed networks are therefore unlikely to encounter interference from competing systems.  Wireless signals are absorbed, reflected, and diffracted by any stalks, trunks, leaves, and branches that obstruct the propagation path. Terrain can also vary considerably, from flat to complex topography depending on the crop type and region, and networks that are deployed in hilly environments must be able to overcome severe shadowing effects. Therefore, the link power budget is dependent on crop growth and terrain in addition to more common factors such as node spacing and antenna height [7]. Therefore, in selecting appropriate wireless spectrum for agricultural networks, careful study of the   42propagation environment must be taken into consideration, as it will greatly affect range and reliability.  3.2.2 Power Consumption  Sensor and actuator nodes in agricultural networks must operate away from convenient sources of power. Therefore, energy consumption is a key design limitation for WSAN nodes. Between sensing, communication, and data processing, a sensor node expends the most energy through data communication. [12]  3.2.3 Range: Node Spacing and Sampling Density  Coverage area requirements are large for farms/fields that often extend over many square kilometres. Agricultural wireless networks are composed of sensors and actuators, and therefore effective communications range must be compatible with the physical range/influence of these devices  Transmission range requirements will depend greatly on the sampling intervals necessary to provide actionable data for specific applications. A number of researchers have studied the underlying range of spatial correlation in soil/climate/plant samples in order to determine the minimum sampling distance required to inform agricultural decision support systems. In [116] the maximum required soil sampling interval for a precision fertilization system was found to be 20m and 60m for two different fields of cereal grains. In [117] the authors conclude from analysis of spatial correlation in paddy field soil measurements that site-specific management of fertilizer application would be possible based on sampling intervals of 20-50m. In [70] a node spacing of 15m was found to be adequate for frost or mildew detection, and detection of fruit maturity could be performed with a node spacing > 15m. In [118] the scale of correlation for different metrics of soil moisture content were found to range from 19.2 ? 94.2m (depending on the particular metric).    43We therefore see that the maximum node separation for actionable agricultural data will range from 15-100m, depending on the quantity of interest, and the variability of the region. Optimal node spacing must also trade-off between the number of messages each node must relay (inversely proportional to spacing) and the energy consumed for each transmission (directly proportional to spacing). Routing simulations for uniformly distributed sensor nodes (in a linear configuration) show an optimal ?hop distance? of roughly 75m. Battery life decays to below 50% of that optimum value when hop distance is < 25m or > 150m [119]. Authors use a ?first-order radio model? with a ?link loss coefficient of 4?. No environment was specified.  The propagation characteristics of the environment will also pose limits on node separation. For example, researchers in [120] have calculated the optimal hop distance for maximum energy efficiency while taking the fading environment into account. The optimal distance was observed to decrease with increasing path loss exponent (from ~135m @ n = 2.4 to ~30m @ n = 3.4) Here, the Rx sensitivity was -94 dBm, the Tx power was 0 dBm, and frequency = 2.4GHz. We therefore see that for a given maximum node separation, dmax (dictated by the geostatistics of the quantity being measured) there is also minimum node separation, dmin that will maximize energy efficiency, which in turn is dictated by the propagation environment. Further, anisotropy in the fading environment (e.g., tree-trunks in forests/groves/plantations, crops planted in rows) will lead to non-uniform propagation characteristics and, therefore, the strongest link may not always be the shortest path  For the uniform node geometries in[45], the next-nearest neighbor for a square grid pattern is ?2 * d and ?3 * d for a triangular grid pattern (where d is the distance to the nearest neighbor). From these three constraints on node separation (sampling density, energy efficiency, and link reliability), we see that maximum effective communications range should be ?3 * min[dmax, dmin ] for triangular node spaceing schemes, and ?2 * min[dmax, dmin ] for square schemes. For the remainder of the paper we will consider the range requirements for agricultural networks to be between 0 ? 200m which covers the use-cases discussed above. To differentiate between use-cases we will consider three   44classes of agricultural networks: class 1 = 0 ? 30m, class 2 = 30 ? 100m, class 3 = 100 ? 200 m.   3.3 Previous Work and its Limitations  3.3.1 Recent Allocation Efforts  Here we seek to summarize the methods and analysis used by other researchers when proposing changes to regional or national spectrum allocations 3.3.1.1 Opening the 433 MHz Sub-band for RFID  In [121], the FCC adopts regulations that increase the maximum allowed field strength and transmission duration to support 433 MHz RFID systems for tracking shipping containers in commercial and industrial areas. Savi Technology, Inc proposed these rule changes for the 433 MHz band in Nov. 2000. They cited better propagation characteristics at 433 MHz and usage of the band by RFID systems in other countries as reasons for pursuing the requested changes.   In response, the FCC proposed changes that would allow: 1) RFID tags to transmit data at average field strength of 11,000 ?V/m (measured at 3m), 2) a peak field strength of 110,000 ?V/m, and 3) transmissions of 120 seconds with at least a 10 second silent period between transmissions. These changes would be permitted provided out-of-band emissions met the limits in Section 15.209. Other companies (Interlogix, Mattel, VYTEK) supported this proposal, stating that the changes would harmonize the band with ITU recommendations and similar systems currently used in Region 1. They also stated that the changes would provide ?considerable savings to manufacturers in producing one model of device for sale in both the United States and Europe would enhance manufacturers? ability to better compete in world markets.? They further stated that such a ?system operates with low power and that operations will be restricted to commercial/industrial environments that severely restrict propagation?   45 A number of parties objected to the proposed changes on the grounds that they would interfere with commercial and residential remote control transmitters, and door and gate controls, though no interference analysis was presented. The AARL was also concerned about interference to amateur radio operations in the band. The National Telecommunications and Information Administration also objected on grounds that the band is allocated to the Federal Government on a primary basis for various radio applications that are essential for homeland defence.   Savi responded by modifying its recommendations and the following changes were agreed to by the FCC. They: 1) set the permissible frequency band of operation to 433.5-434.5 MHz, 2) prohibited operation within 40 kilometers of five Federal Government radar sites 3) required that the locations where 433 MHz RFID systems operate be registered, 4) lowered the peak field strength limit from 110,000 ?V/m to 55,000 ?V/m at 3 meters, 5) reduced the maximum transmission duration from 120 seconds to 60 seconds, 6) adopted a narrower definition of RFID systems, including adding a prohibition on voice transmissions, and 7) restricted operation of RFID systems at 433 MHz to commercial or industrial areas. The FCC concluded that ?the public interest would be served by allowing operation of improved 433 MHz RFID systems.? 3.3.1.2 DSRC Allocations at 5.8 GHz in the Late 1990?s  Allocations in the 5.850-5.925 GHz by the FCC for DSRC systems were made in 1999[122]. The "Notice of Proposed Rule Making" was filed in 1998 [123] and cites a petition made by ITS America in 1997. The ITS America report cites heavily evidence from a report by ARINC in 1996: "Dedicated Short Range Communications (DSRC) 5.8 GHz Band Occupation Considerations". This report is included in the ITS America report as attachment 5 to appendix L (starting at page 276). Quoting at length from the ITS America report: "ARINC employs a relatively simple calculus to determine the minimum channel bandwidth for DSRC. First, the minimum data rate required to   46support each DSRC application is estimated. The results demonstrate that all of the DSRC applications described above can be supported by a data rate of 600 kbps. Next, ARINC combines this data rate with the expected transmission characteristics of DSRC systems under various proposed standards, and concludes that 3.6 MHz of bandwidth supports transmission. ARINC then adjusts this bandwidth to allow for multiple transmissions at different frequencies without causing interference to each other or out-of-band users. Presupposing the use of filters, the channel bandwidth then becomes 5.616 MHz. Finally, ARINC slightly expands this bandwidth to allow for some drift of the center frequency and normal aging and degradation of components. The result is six MHz channels." The details in the report show that ARINC used an analysis of the required bit rate to determine bandwidth. A recommendation of 75 MHz was made to allow for eight such 6MHz channels and to facilitate interoperability with bands in Canada and Mexico. Range requirements dictated transmit power given the propagation characteristics in typical environments. Spectrum occupancy and bandwidth seem to be the main reasons that ARINC recommended 5.8GHz over 915 MHz. They cited heavy use of the 915MHz ISM band to request:  "? a co-primary allocation to permit sharing with the existing services authorized for the 5.850-5.925 GHz band.  Government radiolocation, fixed satellite service uplinks, ISM, amateur radio operators and Part 15 devices are currently authorized to operate in the band." However, they believed that the relatively short links of DSRC caused by high attenuation of signals in the proposed band would allow for high frequency reuse and low interference with these other primary services.  The ARINC report also cited allocation of 10MHz of spectrum in a nearby band (5.795-5.805) by the ERC in region 1 in 1991 [124]. They argued that such harmonization would allow manufacturers to reach economies of scale that would benefit the field as a whole.   473.3.1.3 ITS Allocations at 5 GHz in Europe  In [125] and [126] the European Telecommunications Standards Institute (ETSI) proposed that the Electronic Communications Committee (ECC) of the European Conference of Post and Telecommunications Administrations (CEPT) allocate 2x10 MHz protected bands for ITS services in the 5GHz range. They stated that the available 5 GHz WLAN and 5.8 GHz ISM bands did not provide adequate protection for such high priority, safety-related services and therefore recommended using the 5.885 GHz to facilitate harmonization with FCC regulations. An estimated link budget was presented (using free-space path loss) to determine the minimum EIRP necessary to meet range requirements given the sensitivity of the receiver and the estimated data rate.  The following reasons justify why existing WLAN (IEEE 802.11a and IEEE 802.11h) systems were not sufficient for this application: 1) Sharing the spectrum with users of public WLAN channels causes unacceptable levels of interference to safety critical applications. 2) IEEE 802.11h implements a DFS mechanism to avoid co-channel operation with radar systems. DFS requires to silence all transmissions on a channel to test for the presence of a radar. For time critical safety applications this mechanism is impractical. 3) Safety critical applications require protection and prioritization over public WLAN applications that can allocate channels for several seconds. 4) The WLAN system uses channels with 20-MHz bandwidth. For vehicular-to-vehicular communications a 10 MHz channel bandwidth is used to obtain larger communication ranges and to reduce the symbol interference. The bisection of the bandwidth increases the Guard Interval (GI) to 1.6 ?s which is necessary to cope with the severe multi-path environments expected in highly reflective vehicular environments.   ECC Report 101 [127] uses system parameters defined in the ETSI reports above to analyze compatibility with existing services operating in the 5 GHz band. The Committee uses a two-slope breakpoint, log-distance path-loss model (exponents between 2.7 and 4.3 for Rural, Suburban, and Urban environments) to estimate the interference to adjacent services generated by ITS services. The report concludes that ITS would not suffer   48interference from adjacent services and would not cause interference to such services provided they could tolerate unwanted emissions of -55 to -65 dBm/MHz (depending on the band) 3.3.1.4 Digital Dividend Band for Mobile Communications in Europe  In [127] and [128] the ECC of the CEPT discussed the regulatory aspects of using the digital dividend band (470-862 MHz) for mobile communications services. They summarized existing users of the spectrum in question and assessed the potential for interference with these adjacent systems. Two approaches are compared, namely: 1) allocate the whole band for such services, at the cost of potential global harmonization and antenna efficiency, or 2) allocate a narrowband portion of the spectrum that can be globally harmonized. They concluded that the first approach is possible depending on negotiations between neighboring European administrations.   However, because several European countries had already begun allocations for mobile communications using portions of the digital dividend band (with agreements ranging from 10-15 years), allocating a narrow sub-band which has the potential for global harmonization would not be possible on a mandatory basis within the European region until ~2020. They therefore recommend allocating a harmonized sub-band on a non-mandatory basis. 3.3.1.5 Harmonization Analysis for Wireless Medical Applications  In [129] the author discusses current wireless medical device technologies and the related regional and global regulations and standards. Several ISM, primary, and secondary allocations are available for such devices worldwide. However, the author finds that a number of wireless medical services are not well harmonized. MICS is globally harmonized in the 402-405 MHz band, however the US and Europe have expanded MICS into 401-402 and 405-406 MHz. WMTS is used in 608-614 MHz on a license-exempt basis in Canada and a primary basis in the USA (which also has allocated 1395-1400 and 1429-1432 MHz). However none of these bands have been allocated for WMTS in   49Europe. UWB Radar imaging rules and regulations in the band 3.1-10.6 GHz are not harmonized. The author concludes that regional and global harmonization of these services will users, the wireless medical industry, and regulators and will increase the quality, reliability, delivery, and cost-saving of healthcare.  3.3.2 Carrier Frequency and System Guidelines for Agriculture and WSAN  No spectrum has been specifically allocated to agricultural applications either internationally by the ITU or regionally by national regulators. More broadly however, there has been some discussion of frequencies for wireless sensor and actuator networks for various applications (e.g., [112] and [130])  Unlike urban environments where high frequencies are often selected to minimize the degree of man-made interference, in rural environments it is preferable to use lower frequencies to minimize propagation loss [130] Specifically, for systems requiring large coverage area and low power, the higher portion of VHF and the lower portion of UHF bands are attractive options [112].  According to (EE Times, ?Understand wireless short-range devices for global license-free systems?) [131] the choice available to SRD designers is limited to those portions of the spectrum that allow license-free operation: 13.56 MHz, 40 MHz, 433 MHz, 2.4 GHz, and 5.8 GHz globally; 868 MHz and 915 MHz in Europe, U.S, Canada, Australia, and New Zealand. For systems that require wide range and low power, the sub 1 GHz bands remain compelling. The 433-MHz band (only 2 MHz bandwidth) is one option for global usage, with a slight frequency modification required for Japan. Such customization is ?easily handled by modern frequency-flexible transceivers?. The bands around 868 MHz (Europe) and 902 MHz to 928 MHz (U.S.) are more useful; they do not restrict applications, and they allow more compact antenna implementations (though not globally harmonized). Frequency hopping, duty-cycle limits, and ?listen-before-talk? methods are some of the methods available to mitigate or avoid interference    503.3.3 Carrier Frequency and System Guidelines for Ubiquitous Sensor Networks  Here, we summarize the essential observations and recommendations concerning Ubiquitous Sensor Networks that have been made in key ITU-R reports and Recommendations.   Recommendation ITU-R M.2002: ?Objectives, characteristics and functional requirements of wide-area sensor and/or actuator network (WASN) systems? [112] In Rec. ITU-R M.2002, the authors note that, ?Achieving transmission over long-distances by narrow signal bandwidth is more important than high-speed transmission with broad signal bandwidth.? Further, ?Although a number of frequency bands could be used, considering the propagation characteristics (see [132]  and [133]), man-made noise (see [134]), and the need for a large cell size, it is preferable for WASN systems to use the higher portion of VHF or the lower portion of UHF bands.? The recommendation does not mention agriculture in ?Service applications?, however, it does mention ?environment observation?, ?disaster prevention and measures?, and ?reduction of environmental impact? which could be interpreted in an agricultural context Some mention is made of interference avoidance and mitigation, including a statement that,  ?In order to minimize the potential for co-channel interference, the systems should reduce radiated power from WTs even in their inactive cycle. However, collisions may occur with the increased number of WTs. Due to the bandwidth limitation of the frequency band, some requests from the WTs may not be able to access the system as a consequence of congestion. To effectively accommodate all the WTs in the system, the WASN systems need to employ effective medium-access protocols including priority access schemes.? They note that the system transmission rate for WSAN is between 1-10 kbps and the density of nodes is between 1-1000000 / km2    51Report ITU-R M.2224: ?System design guidelines for wide area sensor and/or actuator network (WASN) systems? [130] Rep. ITU-R M.2224 doesn?t mention agriculture explicitly, though it does mention ?meteorological observation?, and ?environment observation? which could be interpreted in an agricultural context. The authors note that ?The services need to be supported under non-line-of-sight propagation environments for indoor/outdoor communications. In urban areas, in order to avoid high levels of man-made noise, it is preferable to use relatively higher frequencies, since levels of man-made noise tend to be higher at lower frequencies. Conversely, in areas where the noise does not cause serious problem, e.g. rural areas, it is preferable to use lower frequencies to reduce the propagation loss.? The authors recommend special diversity techniques at the base station to mitigate fading. Types of applicable systems include: 1) IEEE 802.15.4 WASN with GSM/EDGE/UMTS/HSPA/LTE backhaul, 2) IEEE 802.16 working group is developing IEEE 802.16p which provides MAC enhancements for devices with low power consumption and burst-y transmissions, 3) IMT-2000 (background-class: [135]) systems can support machine-to-machine (M2M) remote sensor networks, and 4) VHF WASN typical system parameters (example of meter-reading system): cell size = 3.5 km, wireless terminals in a cell = 20000, transmission rate = 9600 bps. They use the Okumura-Hata model describes path loss for suburban environments and consider transmission power ~ 10dBm (up to 30 for urban areas).  Report ITU-R SM.2153-1 ?Technical and operating parameters and spectrum requirements for short range radiocommunication devices? [136] Report ITU-R SM.2153-1 replaces Rec. ITU-R  SM.1538-2. The authors begin by noting that Short Range Devices (SRDs) are radio transmitters which provide either unidirectional or bidirectional communication and which have low capability of causing interference to other radio equipment.  The report doesn?t explicitly mention agriculture, though other LAN technologies and data collection applications are discussed. They note that SRDs are generally not   52permitted to use bands allocated for: radio astronomy, aeronautical mobile, or safety of life services including radionavigation. Global ISM bands available to SRDs include: 6,765-6,795 kHz, 13,553-13,567 kHz, 26,957-27,283 kHz, 40.66-40.70 MHz, 2,400-2,483.5 MHz, 5,725-5,875 MHz, 24-24.25 GHz, 61-61.5 GHz, 122-123 GHz, and 244-246 GHz. Other globally available bands include, 9-135 kHz: Commonly used for inductive short-range radiocommunication applications, 3,155-3,195 kHz: Wireless hearing aids (RR No. 5.116), 402-405 MHz: Ultra low power active medical implants Recommendation ITU-R RS.1346, 5,795-5,805 MHz: Transport information and control systems Recommendation ITU-R M.1453, 5,805-5,815 MHz: Transport information and control systems Recommendation ITU-R M.1453, and 76-77 GHz: Transport information and control system (radar) Recommendation ITU-R M.1452. The report also discusses maximum EIRP for these bands in each region.  Recommendation ITU-R SM.1896 ?Frequency ranges for global or regional harmonization of short-range devices? Rec. ITU-R SM.1896 states that SRDs are not ISM applications as defined in No. 1.15 of the Radio Regulations (RR). It describes same list of globally available bands as in ITU-R SM.2153-1 and mentions other bands that are ?regionally harmonized?, including: 1) 7,400-8,800 kHz: Region 1, Region 2, and some countries in Region 3, 2) 312-315 MHz: Region 2, and some countries in Regions 1 and 3, 3) 433.050-434.790 MHz: Region 1 and some countries in Regions 2 and 3, 4) 862-875 MHz: Region 1 and some countries in Region 3, 5) 875-960 MHz: Region 2 and some countries in Regions 1 and 3.  3.3.4 Models of Path Loss in Vegetated Environments  As noted in Section 3.2.1, when selecting appropriate wireless spectrum for agricultural networks, careful study of the propagation environment must be taken into consideration, as it will greatly affect range and reliability. However, only a few researchers have investigated propagation in agricultural environments:    53In [44], path loss at 433 MHz was measured in a potato field: no model was presented, though the authors noted that path loss increased by 15 dB when the density of vegetation was greatest during the growing cycle.   In [37] the authors compare several models (presented in the following sections) using measurements of path loss for low antenna heights (<0.5m) at 2.4 GHz for environments containing low bushes and ground cover.   In [45], the authors fit a log-distance path loss model to measurements taken at 2.4 GHz in wheat fields throughout several stages of the growing season.   In [46], the authors measured received signal strength at 2.4 GHz in corn fields: no model was developed, though plots of the distance relationship for various antenna heights were presented; the authors noted that the vegetation loss increased for higher density crops, and that received signal strength was highest when the antennas were roughly half the crop height (1.6m in this case).   Accordingly, we have therefore surveyed the literature for models of path loss through vegetation more broadly (i.e., wheat fields, forests, orchards, plantations). 3.3.4.1 Modified Exponential Decay (MED)   The Modified Exponential Decay (MED) model was developed by (among others) Weissberger in the early 1980?s. It applies to the case of propagation through (rather than over) a body of trees. It has the general form below, where A, B, and C are parameters estimated from measured data, f is the carrier frequency, and d is the separation distance between the transmitter and receiver:  The model does not take into account measurement geometry [137]. Different parameter values have been proposed for various scenarios (e.g., frequency, type and density of vegetation, dominant propagation mechanism).    54 Weissberger?s model [138] is based on measurements conducted in several environments (deciduous forests with underbrush, deciduous and pine forests with underbrush, lime tree orchards) at frequencies between 230MHz and 96 GHz. It is given by,  where f is in GHz. The difference between leaves-on and leaves-off state was observed to be about 3-5 dB at 450-950 MHz. This model is suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz in vegetated environments.  The ITU-R (CCIR) model [139] was developed from measurements carried out mainly at UHF. It was proposed for the case where either the transmitting or receiving antenna is near to a small ( < 400 m) grove of trees such that the majority of the signal propagates through the trees [140]  where f is in MHz and d < 400 m. This model is suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz in vegetated environments.  The Fitted ITU-R (FITU-R) model is a new set of parameters for the ITU-R model were proposed in [141]. These parameters were estimated using the method of least squares on  measured data collected at 11.2 and 20 GHz in two environments (a line of deciduous trees, and a wedge-shaped copse of sycamore trees)  where f is in MHz. This model is NOT suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz (developed for much higher frequencies)  The COST235 model is a new set of parameters that were estimated by applying the method of least squares to datasets obtained from measurements in a number of environments (sycamore/lime tree copse, line of horse chestnut trees, pecan forest, and apple orchard) at frequencies between 9.6 ? 57.6 GHz [142]    55 where f is in MHz. This model is NOT suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz (developed for much higher frequencies)  The FITU-R and COST235 models seem to indicate a faster rate of decay of signal strength (with increasing distance) for the case of obstruction by a small number of trees than for the case of obstruction by a large number of trees. This is likely because the dominant propagation mechanism at short distances is the direct path which is strongly attenuated by the vegetation [143]. At larger distances, reflections from nearby vegetative scatterers become the dominant paths [143]. Attenuation is greater in the ?leaves-on? state due to increased absorption , and at higher frequencies when the dimensions of leaves and small branches become large relative to wavelength [143]. 3.3.4.2 Modified Gradient Models  The Maximum Attenuation (MA) model [144] was proposed by the ITU-R for frequencies between 30 MHz and 30 GHz for the case when one terminal is located outside the vegetated area and one terminal is located inside. The model takes the form   where Am is the maximum attenuation and R0 is the initial gradient of the attenuation curve. Some measurements in frequencies between 900 MHz and 1.8 GHz have shown a frequency dependence in Am,   where A1 and ? have been observed to vary with antenna height configuration and vegetation type and density. This model is suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz in vegetated environments.  The Nonzero Gradient (NZG) model is based on the following observation: In the maximum attenuation model, the excess attenuation due to vegetation approaches a   56maximum value as separation distance increases. However, measurements made at other frequencies by [145] and [137] did not display this same phenomena, but showed instead that attenuation continued to increase with increasing distance. The NZG model was proposed in [146] to overcome this ?zero final gradient problem? in the MA model. The NZG model takes the form  where R0 and R? are the initial and final attenuation rates (in dB/m) and k is the final attenuation offset (in dB). This model is NOT suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz (developed for much higher frequencies)  The Dual Gradient (DG) was proposed in[146], the dual gradient model provides correction factors for the NZG model which take into account carrier frequency and antenna beamwidth, i.e.,    where f  is in GHz, and w is the ?maximum effective coupling width? between the TX and RX antennas. However, this model predicts that, all things being equal, path loss will decrease with increasing frequency. As stated in [147] this contradicts both common-sense and measured data (e.g.,[143]) Version 3 of the ITU-R Recommendation P.833 [144] also suggested modifications to the NZG model to account for the effect of operating frequency (recommended for f  > 5 GHz). This model is NOT suitable for this use-case (developed for non-isotropic antennas). 3.3.4.3 Ground and Canopy Reflections  In [140] the authors propose modifying empirical path loss models to account for each of: (1) a single ground-reflected ray, and (2) a single canopy-reflected ray. In each case, for the ray to be significant the distance between the antenna and the reflection surface must   57be within the maximum width of the first Fresnel Zone (for a given operating frequency and separation distance), i.e., . If only the ground reflected ray is significant, the correction can be applied by adding the following to the vegetation loss, . The phase difference between the direct and reflected ray is calculated from . For the case when the canopy-reflected wave is also present, the correction can be applied by adding the following factor to the vegetation loss, . Here, the phase difference for the ground ray is calculated as above (i.e., using the distance from the ground to the antenna), while the phase difference for the canopy ray is calculated using the distance from the antenna to the canopy. The authors found that accounting for both the ground and canopy reflected rays greatly increased the accuracy of the path loss models studied (COST235, ITU-R, Weissberger) for palm and mango plantations at VHF. 3.3.4.4 Log-Distance Model  A number of researchers have found that the log-distance path loss model provides a good fit to measured data in vegetated environments (e.g., [39], [43], [45],[38]):  This model encompasses all losses experienced by the wireless signal (i.e., including free space loss).    58The model has little predictive power for new environments or antenna configurations, therefore the path-loss exponent ?n? and reference value ?L0? must be estimated for each new scenario. This model is not suitable for comparison of path loss at frequencies between 433MHz and 2.4 GHz in vegetated environments (all studies conducted at 2.4 GHz only, with the exception of [38] at 1.8 GHz) 3.3.4.5 Radiative Energy Transfer (RET) Theory   Radiative energy transfer theory was derived from EM theory and requires more detailed knowledge of the environment. It has thus far been underutilized [36]. RET models vegetation as a statistically homogeneous medium of random scatterers and absorbers.   PR is the received power and Pmax is the signal strength received in the absence of vegetation (remainder of parameters described in [36] and [143]). The parameters of the model vary according to vegetation type, leaf condition, and frequency. Parameter values for forested environments (for a variety of tree types) and at frequencies between 1.3 ? 61.5 GHz are given in ITU-R P.833 [144] 3.3.4.6 Comparisons and Goodness of Fit  A number of studies have compared the fit of previously proposed path-loss models to data measured in a variety of environments.  In [140] the COST235, Weissberger, and ITU-R models were compared using measurements of path loss in: 1) a palm plantation wherein trees are 10m tall and spaced in a regular grid ~1.2m apart, and 2) a rainforest wherein trees are ~10m tall and   59irregularly spaced, and there exists a dense underbrush that reaches ~4m high. The frequency bands studied included 40-80 MHz, 250-300MHz, and 500-630 MHz. The COST235 model was found to provide the best fit for the trunk-dominant palm plantation, while the ITU-R model was found to provide the best fit in the dense rainforest. Because of the height of the canopy layer, canopy reflections were only found to be significant at low VHF (~40MHz), while ground reflections were found to be significant up to low UHF (~550 MHz).  In [36] authors compare path loss models using data collected in palm and mango plantations (regular rows, ~5m trees, dense canopy) at 0.4 - 7.2 GHz. Models compared included: Free Space, COST235, FITU-R, ITU-R, Weissberger, MA, NZG, RET. The authors found that, overall, the lack of homogeneity of vegetation density in the environment resulted in significant spatial variation of signal strength (i.e., shadow fading). For the case of propagation through the trees (as opposed to in between the rows), the COST235 model was observed to have the lowest RMSE across all frequencies. For the case of propagation between rows of trees, the COST235 model actually performed quite poorly, and the RET, ITU-R, and Free-Space (adjusted for ground reflections using the method in[140]) provided the lowest RMSE. The MA and Weissberger models were found to perform poorly at all frequencies and for all antenna configurations.  In [148] the authors compare the fit of the NZG and MA models to measurements of path loss at 11.2 and 20GHz through: (1) a line of deciduous trees, and (2) a wedge-shaped copse of sycamore trees. Though both models provided good agreement with the measured data, the NZG model was a better fit at short distances.  In [42], the (Weissberger) MED, ITU-R, and FITU-R models were compared using measurements taken in a plum orchard at 2.4 GHz. The authors found that estimating their own coefficients to the measured data provided a better fit than any of the three previously developed models. The three existing models consistently overestimated path loss by ~10dB. This is likely because those models were developed using measurements   60of path loss in forested environments and the plum orchard pictured in this study has a very low density of vegetation comparatively.  In [37] the authors compare the Weissberger, COST235, FITU-R, and ITU-R models using measurements of path loss for low antenna heights (<0.5m) at 2.4 GHz for environments containing low bushes and ground cover. The authors found that none of the previously developed models provided a good fit to their measured data. This prompted them to modify the models by using two correction factors: (1) the percentage of vegetation in the propagation path, and (2) a specific attenuation value dependent on the type of vegetation. With these new correction factors, the Weissberger model provided the best fit to the measured data.  According to the authors in [147] researchers in [50] (paper not available online) compared the MED, MA, and NZG models using measurements conducted at 1.2, 3 and 11.6 GHz. For all frequencies and antenna configurations considered, the NZG model was observed to provide the best fit to measured data, while the MA model provided the worst fit. In [49] the authors compare the fit of the RET, ITU-R, FITU-R, MED, NZG, and DG models to measurements of path loss in a line of horse chestnut trees (height ~ 14.8 m) at 2 and11.2 GHz. The RET and ITU-R/FITU-R provided the best fit to measured data at both frequencies, while the NZG model provided a poor fit, and the MED model provided the worst fit overall 3.3.4.7 Discussion  Many models have been developed and evaluated for propagation through vegetation, though no clear consensus exists as to which models are more suited to which environments. Also, the majority of these models are unable to account for the large variety of: vegetation density, vegetation morphology (i.e., height and physical extent), and planting geometry (i.e. rows and plant spacing).    613.4 Key Issues for Spectrum Selection 3.4.1 Harmonization  Global or regional harmonization of spectrum allocations for a given wireless service allows manufacturers to achieve larger economies of scale and operators to achieve more rapid rollouts of new services [149]. Recommendations for harmonization efforts include: 1) regulations should not set greater limits than necessary to achieve its goals (e.g., aimed at broad categories of services rather than a specific technology: IMT-2000 standards, vs. UMTS), 2) competing services and technologies should be allowed access to the spectrum, and 3) the creation of standards for harmonized bands should be left open and led by industry.  3.4.2 Path Loss and Propagation Impairments  Path loss determines coverage area/effective communication range. In the case of sensor and actuator networks, this has implications for node spacing and network topology. For a given inter-node separation, dictates required transmit power, and therefore energy consumption/battery life. Accurate prediction of path loss allows for performance prediction, simulation, and fair comparison between network technologies 3.4.2.1 Range Comparison  Here we have calculated the maximum range achievable for several frequency bands (433 MHz, 915 MHz, 2.4 GHz), and using several different path-loss models that apply to signal propagation through vegetation. For the calculations below, we have assumed the following system parameters  ? RX Sensitivity :     S = -80/-90/-100 dBm ? TX Power:     PTX = 0 dBm ? Antenna Gain:   GTX = GRX = 2.15 dBi We can then calculate the maximum achievable range by:  S = PTX + GTX + GRX ? Lfree space - Lveg   62where:  Lfree space = 10 log10 (4?df / c)2 = 20 log10 (d) + 20 log10 (f) ? 147.55 and Lveg is estimated using the models below.  By examining the use-cases of agricultural networks in Section 3.2.3, we have proposed three classes of range requirements (0-30m, 30-100m, 100-200m). The calculations below suggest that, for the shortest range class there may be no benefit to using the 433 MHz band over the 868/915MHz (regionally harmonized) and 2.4GHz (globally harmonized bands), depending on the sensitivity of the receiver. However, as the range requirement increases, several of the models predict that the higher frequency bands will not be able to achieve links of these distances. In Figure 9, Figure 10, and Table 1, we show the range predictions of the above models for the three frequency bands of interest.   63  Figure 9 ? Maximum communications range for -80, -90, and -100 dBm RX sensitivity   Figure 10 ? Maximum communications range for -80, -90, and -100 dBm RX sensitivity (shorter y-axis)     0 1000 200002004006008001000120014001600Frequency (MHz)Max range at RX sensitivity (m)  WeissbergerITURMax. Atten. 1Max. Atten. 2Max. Atten. 30 1000 200002004006008001000120014001600Frequency (MHz)  0 1000 200002004006008001000120014001600Frequency (MHz)  0 1000 20000100200300400500Frequency (MHz)Max range at RX sensitivity (m)  WeissbergerITURMax. Atten. 1Max. Atten. 2Max. Atten. 30 1000 20000100200300400500Frequency (MHz)  0 1000 20000100200300400500Frequency (MHz)    64  Class Model Formula Refs Parameter Values Range @ 433 MHz Range @ 915 MHz Range @  2.4 GHz Modified Exponential Decay (MED) Weissberger 14m < d ? 400m L = A x f BdC  f in MHz (unless otherwise stated) [138] A = 1.33 B = 0.284 C = 0.588 (f in GHz) 195 m 117m 58.4m ITU-R  [139] A = 0.2 B = 0.3 C = 0.6 154m 92.1m 46.3m COST 235 (in-leaf) [142] A = 15.6 B = -0.009 C = 0.26 37.0m 26.1m 16.1m COST 235 (out-of-leaf) [142] A = 26.6 B = -0.2 C = 0.5 26.5m 25.6m 23.0m Fitted ITU-R (in-leaf) [141] A = 0.39 B = 0.39 C = 0.25 356m 146m 44.4m Fitted ITU-R (out-of-leaf) [141] A = 0.37 B = 0.18 C = 0.59 183m 119m 65.4m   65 Class Model Formula Refs Parameter Values Range @ 433 MHz Range @ 915 MHz Range @  2.4 GHz Modified Gradiant Maximum Attenuation L = Am (1 ? exp( -R0 d/Am))  where   Am = A1 f ?  And  R0  ~ 0.08 (@ 433 MHz) R0  ~ 0.19 (@ 915 MHz) R0  ~ 0.5   (@ 2.4 GHz) R0 ~ 1.0  (@ 5.8GHz) f in MHz [144] A1 = 0.18  ? = 0.752  (900-1800 MHz) N/A 155m 48.9m A1 = 1.15  ? = 0.43  (900-2200 MHz) N/A 184m 56.6m A1 = 1.37 ? = 0.42  (16-2118 MHz) 477m 173m 54.9m Non-Zero Gradient L = R?d + k(1 ? exp(-d(R0-R?)/ k))   (f  > 5 GHz) [150] R0  = 4.84dB/m  R? = 0.35dB/m  k = 37.02dB N/A N/A N/A Dual Gradient L = R?d/fawb + k/wc (1 ? exp(-d wc (R0-R?)/ k))  f in GHz [146] (in leaf) a = 0.7 b = 0.81 c = 0.37 k = 68.8 R0  = 16.7 N/A ? For non-isotropic antennas   66 Class Model Formula Refs Parameter Values Range @ 433 MHz Range @ 915 MHz Range @  2.4 GHz R? = 8.77 [146]  (out of leaf) a = 0.64 b = 0.43 c = 0.97 k = 114.7 R0  = 6.59 R? = 3.89 Analytical Radiative Energy Transfer Theory See Refs [143] [144] [36]     Log-distance (Note: these models estimate total path loss, including free space) (measurements @ 2.4 GHz in pine forest) L = L0 +10nlog10(d) + X? [39] L0 = 25.9   n = 3.43 X? = 7.29 dB N/A N/A 98.7m (measurements @ 2.4 GHz in cashew nut plantation) [43] L0 = 50.5   n = 2.28 X? = 4.94 dB N/A N/A 83.4m (measurements @ 2.4 GHz in wheat field, at max veg. dens.) [45] L0 ~ 60dBm (estimated from figure)   n = 2.42 X? = N/A  N/A N/A 26.1m   67 Class Model Formula Refs Parameter Values Range @ 433 MHz Range @ 915 MHz Range @  2.4 GHz (measurements @ 1.8 GHz in  [38](high density) L0 = N/A   n = 4.4 X? = N/A N/A N/A N/A Table 1 ? Path loss models, parameters, and range predictions a 433, 915, and 2400 MHz  68  3.4.3 Spectrum Occupancy  Researchers in the US, China, Singapore, Germany and New Zealand have investigated spectrum occupancy in the VHF and UHF bands and have found that most of the spectrum is heavily underutilized with the exception of broadcast television and cellular services [151]  3.4.4 Interference Avoidance and Mitigation  Neither FCC Title 47 CFR Part 15 [152](in USA) nor RSS 210 [153](in Canada) permit spread-spectrum techniques in the 433 MHz ISM band. Transmissions are limited to < 60s with minimum spacing of 10 seconds. However, further details regarding interference mitigation are not given. Because the majority of agricultural networks are likely to be operated by a single user, and the transmissions are by definition low-power and over short distances, the potential for interference between networks is likely to be small  3.5 Potential Bands  3.5.1 List of ISM Bands:  Current ISM bands that are potentially attractive for use in agricultural applications include: 1) the low VHF ISM bands (e.g., 40.66 ? 40.7 MHz), 2) the sub-gigahertz UHF ISM bands at a) 312 ? 315 MHz in Region 2 and some countries in Regions 1 and 3, b) 433.05 ? 434.79 MHz in Region 1 and some countries in Regions 2 and 3, c) 862 ? 875 MHz in Region 1 and some countries in Region 3, d) 902 ? 928 MHz in Region 2, 3) the high UHF ISM band from 2.400 ? 2.483 GHz, and 4) the low SHF U-NII and ISM bands (e.g., 5.15 ? 5.35 GHz and 5.725 ? 5.875 GHz). Here, we compare the 2.4 GHz, 915/868 MHz and 433 MHz bands.     69 3.5.2 The 2.4 GHz Band  According to the ITU-R Radio Regulations (International)  ? Sec. 5.150 ?2 400-2 500 MHz (centre frequency 2 450 MHz) ? is designated for industrial, scientific and medical (ISM) applications. Radiocommunication services operating within these bands must accept harmful interference which may be caused by these applications. ISM equipment operating in these bands is subject to the provisions of No. 15.13.? In the U.S. and Canada, use of the 2.4 GHz ISM band is regulated by FCC Title 47 Part 5.245, 15.247, 15.249 (USA) and RSS 210 ? Annex 8 (Canada).  Harmonization for agriculture. The 2.45 GHz band is allocated to ISM applications in Regions 1, 2, and 3. Other services that operate in this band include: 1) Fixed (primary user in Regions 1, 2, 3), 2) Mobile (primary user in Regions 1, 2, 3), 3) Amateur (lower half of the band; secondary user in Regions 1, 2, and 3), 4) Radiolocation (primary or secondary user (depending on spec. band portion) in Regions 1, 2, and 3), 5) Mobile-Satellite (space-to-Earth) (primary user in Regions 1, 2, and 3), and 6) Radiodetermination-Satellite (space-to-Earth) (primary user in Regions 1, 2, and 3). Other devices that operate in this band include: Cordless telephones, Baby monitors, WiFi (IEEE 802.11b, 802.11g, and 802.11n), Bluetooth devices, Microwave ovens and ZigBee (and other 802.15.4) devices.  3.5.3 The 915 and 868 MHz Bands  According to ITU-R Radio Regulations (International)  ? Sec. 5.150 ?902-928 MHz in Region 2 (centre frequency 915 MHz) ?designated for industrial, scientific and medical (ISM) applications. Radiocommunication services operating within these bands must accept harmful interference which may be caused by these applications. ISM equipment operating in these bands is subject to the provisions of No. 15.13.?   70 In the U.S. and Canada, use of the 2.4 GHz ISM band is regulated by FCC Title 47 Part 5.245, 15.247, 15.249 (USA) and RSS 210 ? Annex 8 (Canada).  Harmonization for agriculture. The 902-928 MHz band is allocated to ISM applications in Region 2. The 863-870 MHz is allocated to Short Range Devices (SRD) in Region 1. Other services that operate in the 915 MHz band include: 1) Fixed (primary user in Regions 1, 2, and 3), 2) Mobile except aeronautical (primary user in Region 1, secondary in Region 2), 3) Mobile (primary in Region 3), 4) Broadcasting (primary user in Regions 1, and 3), 5) Radiolocation (secondary user in Regions 1, 2, and 3), 6) Amateur (secondary in Region 2). Other devices that operate in the 915 MHz band include: wireless sensor networks, wireless LANs, cordless phones, ZigBee and Bluetooth devices.  Other services that operate in the 868 MHz band include: 1) Fixed (primary user in Regions 1, 2, and 3), 2) Mobile (primary user in Regions 1, 2, and 3) (except aeronautical in Region 1), and Broadcasting (primary user in Regions 1, 2, and 3). Other devices that operate in the 868 MHz band: thermostats, fire systems, burglar systems, wireless microphones and audio systems, home/office automation, and remote controls.  3.5.4 The 433 MHz Band  According to ITU-R Radio Regulations (International), ? Sec. 5.138 ?433.05-434.79 MHz (centre frequency 433.92 MHz) in Region 1 except in the countries mentioned in No. 5.280 ?designated for industrial, scientific and medical (ISM) applications. The use of these frequency bands for ISM applications shall be subject to special authorization by the administration concerned, in agreement with other administrations whose radiocommunication services might be affected. In applying this provision, administrations shall have due regard to the latest relevant ITU-R Recommendations.? ? Sec. 5.280 ?In Germany, Austria, Bosnia and Herzegovina, Croatia, The Former Yugoslav Republic of Macedonia, Liechtenstein, Montenegro, Portugal, Serbia, Slovenia and Switzerland, the band 433.05-434.79 MHz (centre frequency 433.92 MHz) is designated for industrial, scientific and medical (ISM) applications.   71 Radiocommunication services of these countries operating within this band must accept harmful interference which may be caused by these applications. ISM equipment operating in this band is subject to the provisions of No. 15.13. (WRC-07)?  According to FCC Title 47 Part 15.240 (USA)  ? ?Operation under the provisions of this section is restricted to devices that use radio frequency energy to identify the contents of commercial shipping containers. Operations must be limited to commercial and industrial areas such as ports, rail terminals and warehouses. Two-way operation is permitted to interrogate and to load data into devices. Devices operated pursuant to the provisions of this section shall not be used for voice communications.?  According to RSS 210 ? Annex 5(Canada) ? The provisions of this annex are for RFID devices used to identify the contents of commercial shipping containers. Operation must be limited to commercial and industrial areas such as ports, rail terminals and warehouses. Two-way operation is permitted to interrogate and to load data into devices. Voice communication is prohibited.  According to Radio Communications Class Licence 2000 (Australia) [154], ?433.05?434.79 MHz is allocated to Short Range Devices (SRD?s), also called Low Interference Potential Devices (LIPD) with a maximum EIRP of 25mW  According to Radiocommunications Regulations Notice No. 2 (New Zealand) [155], the 433.05?434.79 MHz band is allocated for Short Range Devices (SRD?s), also called Low Interference Potential Devices (LIPD) with a maximum EIRP of -16 dBW.    72 In China, the 432 -438 MHz band is allocated to Radiolocation (primary) and Amateur Radio (secondary) [156] 433 MHz RFID technology (compatible with ISO 18000-7) has also been permitted for use in this band [157].  According to ARIB STD-T67 (Japan) [158], the 426.0375 - 426.1125 and 429.175 - 429.7375 MHz bands are allocated for ?Telemeter, Telecontrol And Data Transmission Radio Equipment For Specified Low-Power Radio Station?  Harmonization for agriculture. The 433 MHz band is allocated to ISM in Region 2 and Active RFID in Canada, USA, Australia, New Zealand, and China. Nearby bands (426 and 429 MHz) are allocated for SRD in Japan. Other services that operate in this band include: Radiolocation (primary user Regions 1, 2, and 3), Amateur (primary user Region 1, secondary user Regions 2, and 3) and Earth exploration Satellite (secondary user Regions 1, 2, and 3). Other devices that operate in this band include: RFID systems, remote keyless entry systems (e.g., automotive), alarms (e.g., fire, security).  Antenna size may still be reasonable despite the relatively long wavelength (? ~ 70 cm.)  3.6 Discussion  In the last 10-15 years, there has been considerable interest in deploying wireless services in agricultural environments to inform decision support systems and facilitate agricultural processes. Agricultural crops, fields, plantations, orchards, and forests/groves contain dense concentrations of vegetation and may experience significant change through the growing season and therefore represent an extreme case for wireless propagation.   For such networks, achieving transmission over long-distances by narrow signal bandwidth is more important than high-speed transmission with broad signal bandwidth, which is fundamentally different from the case of common carrier and personal communication systems where obtaining sufficient throughput/capacity is of primary concern. Coverage area and performance will be determined by the path loss experienced by   73 wireless signals, which is in turn dependent on the nature of the propagation environment and the operating frequency.  A number of path loss models exist for wireless propagation through forested vegetation, however these models are unable to account for the large variety of vegetation density, vegetation morphology (i.e., height and physical extent), and planting geometry (i.e. rows and plant spacing) present in agricultural environments. A few empirical models have been developed for agricultural environments (e.g., wheat and corn fields, tree-fruit orchards), however these models are not suitable for predictions at multiple frequencies. Further work is required to develop models suitable for planning agricultural deployments in a variety of crop types and at different carrier frequencies.  Wireless devices operating in agricultural environments must often operate far from convenient sources of power and therefore have tight energy consumption constraints. Optimal node separation has been shown to increase with path loss, therefore minimizing path loss (e.g., by selecting low-frequency bands) is crucial for extending the lifetime of nodes.   The physical separation of nodes in agricultural networks must be determined by the range of influence of the sensors and actuators that comprise them, and by the sampling density required for the specific agricultural process. By examining the use-cases of agricultural networks, we have observed that node separation requirements for common site specific management applications range from roughly 0-100m (e.g., 20-60m for fertilization systems, 15 for frost and mildew detection, >15 m for fruit (spec. grapes) maturation detection, 20-95m for irrigation). However, to maximize routing options, and to overcome shadowing which may be present in large orchards/plantations, nodes may need to communicate with their next-nearest neighbor (in addition to the nearest neighbor) which, for node separation d is located at ?2 *d for square grid deployments and ?3 *d for triangular grid deployments.     74 Even when the maximum sampling interval (determined by spatial statistics of quantity of interest) may permit denser node deployments, energy consumption constraints may require node separation to be increased (e.g., to reduce the number of messages relayed by each node). We therefore propose 3 range classes for agricultural use cases, namely: (1) 0-30m for applications with high spatial variability (e.g., frost, mildew, and fruit-maturation detection), (2) 30-100m for applications with moderate spatial variability (e.g., some fertilization systems), and (3) 100-200m for applications with low spatial variability (e.g., some irrigation systems)   Current path loss models suggest that, for the shortest range class there may be no benefit to using the 433 MHz sub-band over the 868/915MHz (regionally harmonized) and 2.4GHz (globally harmonized bands). However, path loss predictions show that 868/915 MHz may not be suitable for the longest range class (depending on the sensitivity of the receiver used) and 2.4GHz will likely be inadequate for the two longest range classes.  Since the 868/915 MHz band is not well harmonized, and the 2.4GHz band has insufficient propagation characteristics to meet common use-cases, we therefore propose that the 433 MHz sub-band be opened to permit agricultural wireless services. The 433 MHz sub-band is currently only allocated (e.g., in FCC Title 47 CFR Part 15, and RSS 210) for active RFID services in industrial container and shipping yards, however these environments are unlikely to be co-located with agricultural regions, and the potential for interference between competing services is therefore small. The available bandwidth at 433 MHz is too narrow to permit spread-spectrum interference mitigation techniques, however, because agricultural regions (e.g., farms, fields) are generally operated by a single user, and the devices in question are low-power and operate over short distances, listen-before-talk interference avoidance techniques are probably adequate.   As agricultural deployments begin to scale, it may also be necessary to allocate a nearby band to be dedicated for short-range, low-power, low-throughput agricultural devices. This allocation could have a larger bandwidth to allow for more channels for denser or larger networks. As wireless agricultural services become more diverse, and requirements expand   75 to include higher-throughput communications (e.g., automated vehicle control), a conventional wide-band allocation at higher frequency will be useful.      76 Chapter 4: Interpretation and Implications of the Channel Impulse Response Observed at 2450 MHz in High Density Apple Orchards  4.1 Introduction 4.1.1 Significance of Propagation Environment and Delay Spread   There has been considerable interest in agricultural sensor networks in recent years, however, little work has yet been done to characterize the propagation mechanisms and channel impairments experienced in such environments. A few researchers have studied path loss in orchards (e.g., plum [42] and apple[15]) and fields (e.g., wheat [45] and corn[46]) however, to our knowledge, channel impulse response and delay spread for the case of propagation through vegetation have thus far only been studied in natural forests and woodlands. The degree to which deployments of sensor and actuator networks in agricultural applications are successful will depend on careful study of their specific propagation characteristics.  Delay spread is a fundamental property of wireless communication channels, and describes the multipath richness of the scattering environment. Knowledge of the delay spread in agricultural channels has important implications for signal and equalizer design, which will in-turn affect the performance, reliability, and energy-efficiency of systems and devices. The delay spread of these channels also provides an indication concerning the range over which scatterers that contribute to the response are located, although considerable care must be taken when interpreting these results in multiple scattering environments. In addition, the values of delay spread observed in such environments may influence system deployment strategies especially concerning the height of system nodes and preferred distance from large objects. Finally, the study of the impulse response of these channels may influence the simplifying assumptions made when applying techniques such as vector radiative transfer theory (VRT) and ray tracing to the relatively complicated agricultural environment.     77 4.1.2 Previous Work  Agricultural Environments: To date, measurements of channel impulse response and power delay profile for propagation in vegetated environments have largely been carried out in forests and woodlands. Measurement campaigns have been conducted in natural coniferous (e.g., [159]) and deciduous (e.g., [160]) forests, both of which are fairly homogenous (i.e. the spatial distribution of trees is roughly uniform). By contrast, agricultural environments are often anisotropic, with crops planted at regular intervals in rows. The vegetation canopy is also typically much lower and denser in agricultural environments than in forests and woodlands, and therefore more foliage is likely to be in the path of propagating signals. Additionally, though deciduous forests do transition from a leaves-on state to a leaves-off state throughout the year [161], agricultural environments may experience significantly more change over the growing season such as the growth of stalks and larger structures, in addition to fruit and vegetables.  Scattering Mechanisms and Clusters in Vegetation Environments: In [161], Joshi et al. observed that measurements performed in forests during the leaves-off state resulted in higher estimates of RMS delay spread than with leaves on. In [160], Bultitude observed the same behavior, though for high elevation angle signals where the base station antenna was at 65m above ground and mobile below the forest canopy. Both authors concluded that reflections and scattering are primarily due to tree trunks and branches. Increased foliage density, then, attenuates multipath components, de-weighting the corresponding delay values and reducing RMS-DS. In [162], Ghoraishi et al. utilized a directional channel sounder to observe that the ?forward scattering cluster? contained the greatest amount of radio energy even in the absence of a line-of-sight path. In addition, remaining clusters were observed to come from a narrow range of directions. Oestges et al. [163] also observed that the azimuth spread was fairly consistent at 15 degrees centered on the direct path. Ghoraishi proposed that inhomogeneity in the, ostensibly, homogeneous canopy, acts as a ?canyon? which guides propagating signals towards a group of objects which are near to the receiver and thus have a line-of-sight path to it.   78  Effect of Antenna Height: Joshi et al. [161] observed that RMS delay spread decreased with increasing antenna height for the LOS (i.e., no vegetation in the propagation path) case. However, they did not observe any clear relationship between antenna height and delay spread for propagation through forested vegetation. It is worth noting that all antenna heights studied by Joshi were < 2m above ground and were therefore likely well below the forest canopy. Similarly, Ghoraishi et al. [162] varied the height of transmitting antenna between 4 ? 15m (with the receiving antenna at 3.5m and the average tree height at 10m) and did not observe a clear relationship between RMS delay spread and antenna height.   Effect of Rain: Both Joshi et al. [161] and Meng et al. [164] have observed that RMS delay spread tends to decrease for the case of wet leaves and foliage. Meng et al. [164] further observed that delay spread decreases monotonically with increasing rain rate, from 0 ? 10 mm/hr. However, delay spread was observed to increase during the periods of heaviest rainfall (>10 mm/hr). The authors speculate that as the rain rate increases, more energy is absorbed by obstacles in the propagation path, thereby reducing the contribution of multipath components at the receiver. They further propose that the case of ?heavy? rain is likely to cause movement of the leaves and branches, thereby increasing the number of multipath components and thereby delay spread. Meng et al. also noted that the direct path did not experience increased attenuation with increasing rain rate. The authors propose that this is because this path corresponds to the ?lateral-wave? which propagates above the canopy and is therefore unaffected by rain.  Fading: Oestges et al. [163] observed that delay spread was strongly anti-correlated with ?temporal coherence? (i.e., Rician K factor). Meng et al. [165] captured 1024 instantaneous PDP snapshots at each of their measurement locations in order to assess the temporal variation of specific multipath components. They identified five distinct clusters and observed that the Weibull  parameter, which describes the spread of temporal variation, was strongest for the first cluster (indicating the most coherent contributions), weaker for the second and third clusters, and very low for the remaining clusters (indicating severe fading). Based on the ?four-layer model? (air, canopy, trunk, and ground), the authors concluded that:   79 a. The first cluster was due to the lateral wave propagating along the air-canopy interface, the direct wave traveling in the trunk layer, and the ground reflected wave. This was supported by the high value of the Weibull parameter, which was constant over rain-rate. b. The second cluster was due to the direct wave propagating through the canopy layer. This was supported by the fact that the cluster displayed increasing delay and decreasing Weibull parameter with increasing rain rate, due to increased movement in the canopy layer and increased dielectric permittivity.  c. The third and latter clusters were due to multiple reflections which add increased delay and are susceptible to movement in the canopy layer.  Values of Observed Delay Spread: Ghoraishi et al. (in [162] and [166]) measured values of delay spread ranging from 0.1 ? 15 ns (for various antenna height configurations) for 30dBm, 2.22 GHz signals propagating in a dense forest environment (average tree height = 10m) with antennas approximately 100m apart. Joshi et al. [161] measured values of delay spread from 20 ? 110 ns at distances of 50 to 400 m for 1900 MHz signals in forested environments with antennas mounted less than 2m above ground (i.e., below the forest canopy). Matolak et al. [167] measured delay spread for 33 dBm, 5.12 GHz signals in a moderately dense, forested environment, with antennas at 1.7m above ground. Mean RMS delay spread over distances from 0 ? 120m was observed to be 64 - 87ns while the maximum observed value was 360 ns. Oestges et al. [163] measured delay spreads between 60 to 120 ns for 38 dBm, 1.9 GHz signals in a forest of 3-8m tall deciduous trees at an antenna height of 1.6m and at antenna separations of 40 to 110m. In this case, a directional antenna was used for the transmitter while the receiving antenna was omnidirectional. Meng et al. (in [164] and [165]) measured delay spreads of 132 to 213 ns in a palm plantation (where trees are 5.6m in height and evenly spaced every 7m) for 10W, 240 MHz signals with antennas at 2.15m and an antenna separation of 710 m.    80 4.1.3 Limitations of Previous Work  The studies conducted to date in natural environments (i.e., forests and woodlands) are insufficient to anticipate the channel impulse response of anisotropic agricultural environments with low, dense vegetation canopies. The purpose of this paper will be to study the effect of these artificial (i.e., evenly spaced rows of densely planted crops/trees) environments on the statistics of channel impulse response, and to discuss the implications for design and deployment of agricultural networks. 4.1.4 Objectives  Here we have conducted a measurement campaign to investigate the channel impulse response and RMS delay spread for 915 MHz and 2.4 GHz signals propagating in agricultural environments (spec. an apple orchard). We seek to answer the following questions:   a. What is the range of delay spread values observed in these environments? (implications for signal design) b. Is the frequency selectivity of the channel sufficiently flat to eliminate the need for equalizers and thereby reduce the power consumption of wireless devices? c. What is the relationship between the channel impulse response and antenna separation? d. How does delay spread change when the transmitting antenna is elevated above the vegetation canopy?  e. Is there a relationship between delay spread and path loss? f. How does the channel impulse response for along-row propagation compare to that of cross-row propagation?      81 4.1.5 Outline  The remainder of this chapter will be structured as follows: ? In section 2 we will describe the environment and propagation mechanisms under study. ? In section 3 we will describe the measurement equipment and procedure. ? In section 4 we will present the details of our data reduction procedure and the results of the measurement campaign. ? In section 5 we will discuss implications of the results presented in section 4. ? In section 6 we will present our conclusions.  4.2 Propagation Environment 4.2.1 Details of the Environment Studied  Our measurements were performed in a high-density apple orchard near Keremeos, British Columbia. A high-density orchard differs from a traditional orchard in that individual branches from conventional apple trees are planted directly in the ground so that they grow vertically. This reduces the ?trunk? length and the outward growth of branches, and therefore concentrates fruit production close to the central structure, increasing the growing efficiency of each ?tree?.  In the Keremeos Orchard, we observed three categories of vegetation density, namely: v-shaped, high-vertical, and low-vertical. In the v-shaped case, two of the branches described above were planted at an angle to create a V, and therefore make more efficient use of the open space between rows. In the high-vertical and low-vertical case, only a single branch was planted vertically. The difference in density between the two vertical configurations was the result of the age of the branches, with young branches being thin and sparsely vegetated and older branches having a dense coverage of leaves and fruit. We chose to perform our measurements in the densest (i.e., V-shaped) case as there were more of these rows in the orchard, which allowed us to reach antenna separations of 100 m in all directions.    82 Trees in the V-shaped rows had an average height of 3m, with a few trees having branches that extended to 4m. Rows were an average of 3.5m apart, though one row in the measurement region had a spacing of 6m. In the V-shaped rows studied, foliage extended 1-1.5m from the center of the row (at the tallest point of the trees) in both directions. There was therefore less than 0.5m of free space between rows at the tallest point, and ~3.5m at the base. The ends of each row were supported by a V-shaped wooden structure, and thin metal cables ran along the length of rows to support branches. No other objects were present in the measurement region.  The measurement region was sufficiently long for measurements up to 150m in the along-row direction, but only wide enough for antenna separations up to 100m in the cross-row direction. We therefore chose to limit our measurements to 100m in both directions.     (a) (b)  Figure 11 ? (a) ?High-vertical Rows? consist of single apple tree branches planted at regular intervals. (b) ?V-shaped rows? consist of two branches per row planted at an angle to maximize tree density.  4.2.2 Propagation Mechanisms  For any wireless system operating in a sufficiently cluttered environment, there exist multiple paths that a signal may take between the transmitting and receiving antennas. In addition to   the direct path (i.e., the shortest path), when there is linethe receiver, there are often indirect paths due to reflections from nearby objects. The signal that arrives at the receiver is therefore a copies of the original signal.  Here we consider the case where sensor nodes are deployed within the canopy of a high density apple orchard. In this case, reflections occur off of trees in neighboring rows (see propagation through forested environments, others (e.g., layer? model consisting of the air, canopy, trunk, and ground layers. waves that propagate due to the interaction with these layers: (1) the lateral wave along the air-canopy interface, wave traveling in the trunk layer, andconsidered in this paper (see ground, thereby eliminating the ?trunk layer?. There are therefore three multipath components in the vertical dimension in this scenario (see does not take into account scatterers in the horiz (a)   Figure 12 ? (a) Multipath components in an apple orchard. (b) Interaction between   -of-sight between the transmitter and superposition of multiple attenuated and delayed the direct path is obstructed by a line of trees, anFigure 12a). [165]) have considThere are then four (2) the direct wave traveling in the canopy layer, (3)  (4) the ground reflected wave. For the environment Figure 11), the leaves and branches extend nearly to the Figure 12b). Note that this model ontal plane.  (b) the layers of the propagation environment.   TXLateral WaveGround Reflected WaveDirect Wave83 d For the case of ered a ?four-propagating the direct  RX  84 The delay of the multipath components due to the lateral wave, direct wave, and ground reflected wave will be proportional to their path lengths. For the case of the direct wave, the path length is merely d, the separation between the transmitting and receiving antennas. For the case of the ground reflect wave, the path length is,    	  2 	  2 (1)  where hTX is the height of the TX antenna and hRX is the height of the receiving antenna. For the case of the lateral wave, the path length is,                      (2)  where  is the average tree height,  is half the vertical 3dB beamwidth of the transmitting antenna, and  is half the vertical 3dB beamwidth of the receiving antenna. Similarly, the delay associated with reflections from neighbouring rows can also be calculated using (1) by setting      , where  is the inter-row spacing. In each case the expected delay associated with each path length ! is ! ? $ where c is the speed of light (3.0 x 108 m/s).   4.2.3 Principle of Measurement  Here we have used a vector network analyzer (VNA) to measure the complex frequency response of the channel over a given frequency span. After suitable trace averaging and windowing, we can take the inverse Fourier transform of this frequency response to obtain the channel impulse response. The frequency step size and sweep bandwidth will determine the maximum measurable delay and time-domain resolution, respectively, and therefore must be chosen carefully.  In order to measure the delay of specific multipath components, the time-domain resolution must be less than the delay associated with the lengths of their paths. Using (1),   85 with     2 m and   20 m, the length of the path taken by the ground reflected ray is   20.396 m. The delay associated with the direct path is then 67 ns, while the delay associated with the ground reflected wave is 68.32 ns. In order to differentiate the two paths, the time-domain resolution must therefore be * 1.32 ns which requires a sweep bandwidth of 757.4 MHz in the frequency-domain.   Alternatively, if we wish to identify the multipath components due to reflections from neighbouring rows (for   20 m and a row spacing of 3 m), we must be able to measure delays at .  69.6 ns, .  77.7 ns, and .0  89.7 ns for path lengths   20.881 m,   23.324 m, and 0  26.907 m (for the three nearest rows). In order to resolve these row-reflected paths, our sweep bandwidth must be > 384.6 MHz. We have chosen to use a sweep bandwidth of 1000 MHz centered at 2.5 GHz to give us a 1 ns resolution in the time domain. Since the number of samples is fixed by the VNA at 551 (see Section 4.3.1), the maximum measureable delay is 551 ns which corresponds to a path length of 165m. This allows us to accurately measure the delay spread at our maximum distance of 100m.   4.2.4 Shape of Impulse Response  Erceg et. al [168] observed that the power-delay profile when using directional antennas in outdoor suburban environments follows a ?spike plus exponential? shape composed of a strong return (the ?spike?) at the lowest delay, plus a set of returns whose mean powers decay exponentially (or linearly on a logarithmic scale) with delay (See Figure 13). They characterize this shape by using two parameters, namely, the ratio of the average powers in the ?spike? and ?exponential? components, 34, and the decay time constant (.4) of the ?exponential? component.  The power delay profile is therefore given by,  5.  67.  89:;!?= =>? 7.  ?.?!@4  (3)   where A, B, and .4 are estimated from the data and ?. is the resolution of the measured impulse response. 67. is then the ?main? component and the remainder is the   86 ?exponential? component. 34 can be estimated by determining B, and A from the data and using,   6  34 1  34;  8  1  :;?= =>? /1  34?   . (4)   Erceg et al. further noted that as ?. decreases, .4 approaches the delay spread of the exponential component of the delay profile and is a reasonable approximation even when ?. is large.   The ?spike plus exponential? shape was not observed when using omnidirectional antennas, during which the absence of sidelobes in the antenna patter resulted in a stronger dependence on the presence (or lack thereof) of individual scatters in the vicinity of the antennas. However, since the environment considered here has a much denser, more uniform distribution of vegetation, we anticipate that this shape may provide a good fit to the data obtained in our measurement campaign.  Figure 13 ? ?Spike plus exponential? delay profile shape  4.3 Measurement Configuration and Procedure  4.3.1 Equipment  Our measurement apparatus consists of an Anritsu MS2034A ?VNA Master? vector network analyzer configured to measure S21 by transmitting and receiving a pulse over a given frequency sweep bandwidth. To perform such measurements over distances of 10?s of meters without incurring very high levels of cable loss and unwanted phase reflections, we have ?  (?s) P    87 used MITEQ RF-over-fibre transceivers (see Figure 14). These transceivers convert an RF signal to a modulated optical signal which is then transmitted over fibre-optic cable before being converted back to RF. Power for the measurements was supplied to both the transmitter and receiver equipment by a van equipped with deep-cell storage batteries. We then ran power cables the length of the measurement region to provide uninterrupted power throughout the campaign.    Figure 14 ? Measurement Configuration  VNA RF to Fiber Fiber to RF RX TX PA   88 We have chosen to use ultra wideband (UWB) omnidirectional antennas which allow us to achieve the desired frequency sweep bandwidth of 1000 MHz (at 2.5GHz center frequency) necessary to identify the shape of the channel impulse response (see Section 2). Antennas were mounted within (height = 2m) and above (height = 4m) the tree canopy by using tripods and PVC antenna masts stiffened by inserting a long wooden dowel into the center of the PVC. The receiving antenna, VNA, and RF-over-fiber transmitter were placed in a cart and were stationary over the course of the measurements campaign. The transmitting antenna, PA, and RF-over-fiber receiver were placed on a wheeled dolly which allowed us to vary the antenna separation for each measurement location.   Measurement locations were spaced using a Disto laser range-finder and 100m measuring tape. To assess whether delay-spread has a power-law relationship with distance, we chose our measurement locations to have uniform spacing on a log scale. A 26 dB power amplifier was used to supply 24 dBm of power to the TX antenna which allowed us to reach antenna separations of ~100m. Measurements were automated by use of a laptop running MATLAB which sent SCPI instructions using VISA over TCP/IP. Each sweep consisted of 551 data points while the frequency resolution was determined by the sweep bandwidth. The sweep duration varied but was consistently between 0.5 and 1.5 s. At each measurement location we performed 50 sweeps so that we could average the responses and thereby reduce the noise floor (i.e., increase the signal-to-noise ratio).   4.3.2 Calibration and Validation  Before each measurement run (and periodically throughout the day) we configured the VNA to use the spectrum analyzer mode and swept the bandwidth under study to assess whether any interferers were present. At no point during the measurement campaign did we observe interfering signals in the band of interest.   In addition, before conducting measurements, we removed both antennas and performed a through-line calibration of the measurement apparatus to remove any frequency selectivity and delays introduced by the equipment. In between measurement runs we would again connect the two co-ax lines that feed the antennas and verify that the calibration   89 yielded a flat frequency response with constant phase over the sweep bandwidth. If there was any deviation from the calibration sweep (e.g., because of a change in cable position), we performed a new calibration sweep before continuing.  We measured the gain of our apparatus by removing the two antennas, connecting the feed lines to each other, and measuring C via the VNA. Our system gain was thus observed to be ~24 dB. Therefore, because the maximum input power on the VNA is 23 dBm, the ?high-power? (0 dBm) VNA setting that was used for measurements (necessary to obtain sufficient range) could not be used for calibration. The ?low-power? (-35 dBm) VNA setting was therefore used during through-line calibration, while the high-power setting was used during measurement.   4.3.3 Measurement Configurations  Here we wish to investigate the effects of: 1) antenna separation, 2) transmitter height, and 3) propagation direction, on multipath propagation. We have therefore performed measurements in the following configurations:      Run Relationship Studied1 TX-RX antenna separation  2 Transmitter height3 Propagation directionTable  4.4 Results and Reductions 4.4.1 Data Processing  At each measurement location, we captured 50 snapshots of the complex frequency response (CFR), H(f) , of the wireless channel (see the frequency response is equivalent width equal to the sweep bandwidth. This is equivalent to convolving the underlying channel  Details ? Sweeps measured with stationary RX; TX moved to 15 different locations between 5100m from RX ? TX and RX antennas at 2m above ground (i.e., within canopy) ? Antennas in-line with trees (i.e., direct path is obstructed by row). ? Same vegetation density throughout? Center frequency = 2.5 GHz ? Sweep BW = 1 GHz  ? Same as 1 but with TX antenna at 4m (~1m above canopy)  ? Same as 1 but with TX moved perpendicular to rows (i.e., ?crossSee Figure 15. 2 ? Measurement configurations   Figure 15 ? Propagation Directions  Figure 16). The act of measuring a finite portion of to multiplying the full CFR by a rectangular window of 90 - -row?)   91 impulse response, h(?), with a sinc function in the time domain. The sidelobes of a sinc function are -13.3 dB down from the peak value, and therefore introduce inaccuracies in our estimates of CIR. We have therefore chosen to apply a Blackman window to each snapshot of CFR, which has much lower sidelobes (-95 dB) and thereby reduces the windowing effect of measurement on the time domain response. Our windowed frequency response is then,  DEF  DF G 8F (5)   where 8F is an N-point Blackman window, and N is the number of samples in the frequency domain (551 when using the Anritsu MS2042A VNA Master). This windowing procedure, however, causes a loss of resolution in the time domain, as the width of the main lobe of the inverse Fourier transform (IFT) of the Blackman window is ~ 5 ns (for the 1 GHz span used here).  Figure 16 ? A single snapshot of the channel frequency response (magnitude and phase) for along-row propagation, high TX antenna height, and 6.3m antenna separation.  We can interpret the pass-band frequency response (i.e., the measured, finite portion) as the baseband frequency response by shifting in frequency so that the sweep bandwidth is from 0 ? 1 GHz. Before converting to the time domain, we create a conjugate symmetric series such that  2000 2200 2400 2600 2800 3000-90-85-80-75-70-65Magnitude of Complex Frequency ResponseFrequency (MHz)|H(f)| (dB)2000 2200 2400 2600 2800 3000-180-135-90-4504590135180Phase of Complex Frequency Response? H(f)Frequency (MHz)  92 DEH F  DEF . (6)   This baseband frequency response now has 2N samples. To get the channel impulse response, we then take each baseband frequency response and convert it to the time domain by taking the inverse Fourier transform.  .  IJJKLDEFM . (7)   Because our complex frequency response is conjugate symmetric, we know that our channel impulse response is real-valued. The power-delay profile (PDP) is obtained by taking the square of the CIR and normalizing so that it has unit area,  N.  .? ?.P!@4  (8)   where ?. is the time-domain resolution of our measurement and is equal to the inverse of the span of the frequency response. Here, ?.  1 1GHz  1ns? . We can then average the 50 snapshots of the power-delay profile to smooth out variations due to noise.  An example of the averaged, normalized PDP for the same location as in Figure 16 can be seen in Figure 17, where the quantity on the x-axis is ?excess delay? or the difference between the measured delay and the delay associated with the LOS propagation path. Note that the PDP in Figure 17 has been calculated without any calibration applied to the complex frequency response. The strongest multipath arrives much later than the LOS propagation delay (0 ns in Figure 17). This is due to the time taken for the signal to propagate through co-axial cables of the measurement apparatus. In addition, we observe echoes at -40 dB from the peak that arrive before the shortest possible path (co-ax cable length plus antenna separation) which should not be possible. We observed that these echoes were present for all measurement locations, and have deduced that they are likely due to leakage caused by poor isolation between the two RF ports of the VNA.     93  Figure 17 ? Un-calibrated PDP for along-row propagation, high TX antenna height, and 6.3m antenna separation.   To properly shift the PDP (i.e., remove the delay associated with the cables) and remove what appear to be spurious echoes (see for example the -50 dB multipath components at ~480 ns and 510 ns), we must calibrate out the effects of the measurement apparatus from the measured samples. The measured frequency response, DTF, is  DTF  DUVUF G DWXYZF (9)  where DUVUF is the frequency response of our measurement system, and DWXYZF is that of the wireless channel. Therefore, to remove the effects of the measurement apparatus, we must divide (in linear units) our measured response by the through-line C sweeps we obtained before each measurement run (see Section 4.3.2). 0 100 200 300 400 500-70-60-50-40-30-20-100Power Delay Profile? [ns]p(?) [dB]  94  Figure 18 ? Calibrated PDP for along-row propagation, high TX antenna height, and 6.3m antenna separation.  Applying calibration and performing (5) - (8) to the series shown in Figure 17 results in the delay profile shown in Figure 18. Here we see that the suspected spurious delays have been removed and the PDP has been shifted to remove the delay introduced by the measurement apparatus. The strongest multipath component is now at 0 ns excess delay, which is the propagation delay of the direct path.   Note that we still see the echoes suspected to be due to leakage, though they are now shifted in the time domain by the calibration procedure (now at ~500 ns). If the spur was caused by the measurement apparatus, then it should have been removed by the calibration procedure. It still remains, however, because the calibration was performed using the low power setting on the VNA (see Section 4.3.2) which did not produce these spurious echoes, while the measurements were performed using the high power mode in order to achieve our desired range. We have therefore chosen to ?gate? out these spurs by excluding all multipath components at delays greater than 480 ns from later calculations (all spurs were observed to 0 100 200 300 400 500-70-60-50-40-30-20-100Power Delay Profile? [ns]p(?) [dB]  95 be above this value). This corresponds to removing all multipaths greater than 144m in length, which is significantly longer than the maximum range of 100m considered here. We therefore determined that excluding these samples would not interfere with our analysis. The final windowed, calibrated, and gated power-delay profile is shown in Figure 19.  Figure 19 ? Processed power delay profile for along-row propagation, high TX antenna height, and 6.3m antenna separation.  4.4.2 RMS Delay Spread 4.4.2.1 Moment-Method  The standard measure of multipath richness of a wireless channel is the root mean square (RMS) delay spread, which can be estimated from the second central moment of the power-delay profile [168]: .TU 9.[N[  \9.[N[[ ][ . (10)  0 100 200 300 400-70-60-50-40-30-20-100Power Delay Profile? [ns]p(?) [dB]  96  Before estimating the RMS delay spread, however, the PDP must first be processed to remove samples due to random noise.  Some researchers have chosen to threshold the PDP (i.e., exclude all values below some level) by calculating the statistics of the noise floor and setting a threshold level above the mean noise value (e.g., 6 dB in[168]). Others have chosen absolute thresholds relative to the peak PDP value for excluding echoes due to noise or weak multipath components (e.g., -25 dB in [167] and[164], -40 dB in [161] and [163]). Here, we observed echoes at approximately -40 dB from the peak value at delays earlier than the LOS propagation path (see Figure 19) which must either be spurious components introduced by the measurement apparatus but not removed by calibration, or the effects of windowing in the frequency domain. We have therefore chosen a threshold value of -35 dB from the peak value. However, for the case when the peak echo is less than 35 dB above the noise floor, we have chosen to use samples of the noise floor to estimate the mean (^Z) and standard deviation (_Z).  The threshold in this case is chosen to be 5` XUXabc  ^Z  3_Z, which should remove 99.9% of noise values, assuming they are Gaussian distributed.  We can see the effects of choosing an appropriate threshold value in Figure 20, where our selection of -35 dB from the peak value and using (10) results in an estimate of .TU 4.65 ns. In addition, following [168] we have excluded from subsequent analysis measurements for which the strongest echo was less than 15 dB above the noise floor.   97  Figure 20 ? The effect of threshold selection on estimates of RMS delay spread   4.4.2.2 Slope-Method  As described in Section 4.2.4, Erceg et al. [168] found that plots of power-delay profile from measurements performed in suburban environments at 1.9 GHz with directional antennas followed a ?spike plus exponential? shape. The delay profile, therefore, could be expressed as, N.  67.  89:;!?= =>? 7.  ?.?!@4  (11)  where ?. was the sampling resolution in the time domain, and 6, 8, and .4 could be estimated from data. On a decibel scale, the slope of the ?exponential? portion of the delay profile is therefore proportional to 1 .4? . The decay time constant, .4, can then be interpreted as the ?intrinsic? delay spread of the scattered component of the received signal, or the delay spread that would be measured in the absence of noise (i.e., infinite SNR).  Before estimating .4 from measured data, we must first account for the ?spreading? or loss of time-domain resolution due to windowing in the frequency domain. Following [168], we 0 100 200 300 400-70-60-50-40-30-20-100Power Delay Profile? [ns]p(?) [dB]0 5 10-70-60-50-40-30-20-100?? vs. Noise Threshold (??=4.6482 ns)Noise Threshold (dB)?RMS (ns)  98 have therefore chosen to ?bin? the thresholded values of the delay profile into 4 ns wide bins, which is the width of main lobe of the Blackman window at -35 dB from the peak. The resulting delay profile is shown in Figure 21, where .4 is then estimated from fitting a regression line to the exponential portion of the delay profile. We can see that at close range (6.3m antenna separation in Figure 21) the first two echoes are quite strong, at -3 dB and -7 dB from the initial spike, respectively, while the strength of later echoes decays linearly with time. As antenna separation increases (e.g., 125m in Figure 22) we see a more shallow decay of multipath components with time (i.e., larger delay spread).    99  Figure 21 ? ?Spike + Exponential? fit to power-delay profile for along-row propagation, high  TX antenna height, and 6.3 m antenna separation.  Figure 22 - ?Spike + Exponential? fit to power-delay profile for along-row propagation, low  TX antenna height, and 100 m antenna separation. -10 0 10 20 30 40 50 60-40-35-30-25-20-15-10-505Spike + Exponential Delay ProfileExcess Delay [ns]P( ?) [dB]  Regression: ?0 = 1.7635SpikeExponential-10 0 10 20 30 40 50 60-40-35-30-25-20-15-10-505Spike + Exponential Delay ProfileExcess Delay [ns]P( ?) [dB]  Regression: ?0 = 3.0932SpikeExponential  100  4.4.3 Coherence Bandwidth  A common metric of the frequency selectivity of a wireless channel is the coherence bandwidth, which is a statistical measure of the range over which the fading experienced by two signals transmitted at different frequencies is strongly correlated. We can calculate the coherence bandwidth by using the frequency autocorrelation function, which can be obtained by taking the Fourier transform of the power delay profile [169]:  e?F  f N.:;[g?h=.?;?  (12)  where ?F is the frequency separation. The coherence bandwidth 8$ at a correlation level i is then the minimum frequency separation for which |e8$| k i.   When the bandwidth of a transmitted signal is greater than the coherence bandwidth, different frequency components of the signal will experience different levels of fading, and the channel is said to be frequency selective. It is common to report the coherence bandwidth for a correlation level of 0.9 [169][170][171]. The normalized frequency autocorrelation function at the location considered in Figure 16-11 can be seen in Figure 23. The coherence bandwidth at a correlation level of 0.9 at this location is 10.3 MHz.  Therefore, we could say that any system which has a bandwidth < 10.3 MHz would not experience frequency selective fading in this channel (for an antenna separation of 6.3m).   101 Figure 23 ? Coherence bandwidth for correlation level of 0.9 for along-row propagation, low  TX antenna height, and 6.3 m antenna separation.  4.4.4 Path Loss  The path loss experienced at each measurement location can be estimated by averaging the magnitude of the complex frequency response over the sweep bandwidth  5l  P? D?FP!@4  . (13)   The measured path loss for the four scenarios considered in this campaign can be seen in Figure 24. Here, we have plotted path loss vs. antenna separation on a log-log scale, such that the relationship can be expressed as:  5l   H 10log4  5l4 (14)   0 20 40 60 80 10000.10.20.30.40.50.60.70.80.91Frequench Autocorrelation, Bc0.9 = 10.292 MHzFrequency Separation [MHz]| ?( ?f)|  102where d is in meters, 5l is in dB, and 5l4 is the y-intercept. We have also measured the strength of the relationship between the two quantities by calculating the correlation coefficient, p.  Using regression analysis to estimate the best fit parameters for each data series, we obtain:  5l  4.18 H 10log4  45.8 , p  0.95 (15)  5l  3.18 H 10log4  48.1 , p  0.97 (16)  5l  2.12 H 10log4  61.1 , p  0.95 (17)  5l  2.83 H 10log4  49.6 , p  0.99 (18)   where (15) corresponds to low TX cross-row, (16) corresponds to high TX cross-row, (17) corresponds to low TX along-row, and (18) corresponds to high TX along-row. All four path loss series show very strong correlation (p r 0.9) with distance. The path loss exponent, n is greatest for low-TX height and cross row propagation. In general, we observed that cross-row propagation experiences greater levels of path loss than for along-row propagation. The noise floor of the VNA for these measurements corresponds to a path loss of 115 dB, and therefore values for the low-TX, cross-row case were excluded from this analysis for distances > 50m as the measured path loss for these cases was < 3dB above the noise floor.    103 Figure 24 ? Path loss vs. antenna separation and linear regression    4.4.5 RMS Delay Spread and Distance  In Figure 25 we have plotted .`TU, calculated using the moment-method in (10), vs. antenna separation. We have used linear regression analysis to estimate the best fit parameters and the correlation coefficient to measure the strength of the relationship between the two variables. The solid lines in Figure 25 are the best fit curves given by:  .TU  0.201 H   5.52, p  0.82 (19)  .TU  0.0459 H   5.09 , p  0.39 (20)  .TU  0.0124 H   4.94 , p  0.23 (21)  .TU  0.00324 H   5.21 , p  0.08 (22)   where (19) corresponds to low TX cross-row, (20) corresponds to high TX cross-row, (21) corresponds to low TX along-row, and (22) corresponds to high TX along-row. Here we see that only for the low-TX height, cross-row case is the RMS delay spread strongly correlated with distance. For the high-TX, along-row case, RMS delay spread is approximately uncorrelated with distance and, for the other two cases, only weekly correlated with distance. 4.0 6.3 10.0 15.8 25.1 39.8 63.1 100.060708090100110120Distance [m]Path Loss [dB]Path Loss vs. Antenna Separation  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-Row  104 In Figure 26, however, we see moderate to strong correlation between .4(calculated via the slope method in 4.4.2.2) and distance for all four cases. The best-fit curves in Figure 26 are given by  .4  0.165 H   1.10 , p  0.87 (23)  .4  0.0343 H   2.35 , p  0.58 (24)  .4  0.00942 H   1.84 , p  0.44 (25)  .4  0.0126 H   1.86 , p  0.77 (26)   Here, we still see very little change in .4 with distance (i.e., very small slopes of the linear regression lines) for the along-row case, over the range of antenna separations measured. It is only for cross-row propagation, especially with low TX height, that we see a significant change in .4 at these distances.   105 Figure 25 ? Moment-method RMS delay spread estimates vs. antenna separation.  Figure 26 ? Slope-method RMS delay spread estimates vs. antenna separation.   0 20 40 60 80 100 120051015202530Distance [m]? rms [ns]?rms (Moment-Method) vs. Distance  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-Row0 20 40 60 80 100 1200510152025Distance [m]? 0 [ns]?0 vs. Distance  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-Row  1064.4.6 RMS Delay Spread and Path Loss  In Figure 27 we have plotted .`TU, calculated using the moment-method in (10), vs. path loss as calculated in Section 4.4.4. We have used linear regression analysis to estimate the best fit parameters and the correlation coefficient to measure the strength of the relationship between the two variables. The solid lines in Figure 27 are the best fit curves given by:  .TU  0.106 H 5l  2.08, p  0.38 (27)  .TU  0.189 H 5l  10.4 , p  0.68 (28)  .TU  0.00569 H 5l  5.03 , p  0.03 (29)  .TU  0.00610 H 5l  4.56 , p  0.06 (30)      107Here we see that .TU is approximately uncorrelated with path loss for the along-row case, and weakly to moderately correlated for the cross-row case. The residuals from the regression analysis are especially large (i.e., the fit is particularly poor), for the cross-row case for large path loss (greater than 100 dB).   Figure 27 ? Moment-method RMS delay spread vs. path loss.  In Figure 28, we see that the level of correlation between .4 (calculated via the slope-method in Section 4.4.2.2) and path loss is higher than for the moment-method in all cases. The best fit curves are given by:  .TU  0.158 H 5l  10.7, p  0.83 (31)  .TU  0.108 H 5l  6.32 , p  0.84 (32)  .TU  0.0404 H 5l  1.47 , p  0.59 (33)  .TU  0.0305 H 5l  0.39 , p  0.75 (34)   As with distance, we see that only for the cross-row case do we see a significant change in delay spread with path loss. For the along-row case, delay spread calculated via the slope-60 70 80 90 100 110 1200246810121416Path Loss [dB]? rms [ns]?rms vs. Path Loss  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-Row  108method appears to be approximately constant at 2-3 ns over the range of path loss values observed.  Figure 28 ? Slope-method RMS delay spread vs. path loss. 4.4.7 Coherence Bandwidth and RMS Delay Spread.  Researchers have observed previously (e.g.,[171]) that the coherence bandwidth at a correlation level of 0.9 follows an inverse relationship with RMS delay spread, and can be expressed by:  8$4.s  t.TU . (35)   Yang et al. [171] proposed a value of t  0.63 for 60 GHz indoor channels, while Cox et al. [172] proposed a value of t  0.9 for 910 MHz outdoor channels. In Figure 29 we see that the value proposed by Cox (t  0.9) provides a good fit to the moment-method estimates of RMS delay spread, though it over-predicts for high .TU (i.e., greater than 10 ns). In Figure 30 we see that the slope-method estimates of delay spread do not fit the inverse relationship in (35) as well as the moment-method estimates. 60 70 80 90 100 110 120-2024681012Path Loss [dB]? 0 [ns]?0 vs. Path Loss  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-Row  109 Figure 29 ? Coherence bandwidth vs. moment-method RMS delay spread.  Figure 30 ? Coherence bandwidth vs. slope-method RMS delay spread   0 5 10 15 20-1001020304050607080?rms [ns]Bc 0.9 [MHz]Coherence Bandwidth vs. RMS Delay Spread  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-RowBc0.9=0.9/?rmsBc0.9=0.63/?rms0 5 10 15 20-1001020304050607080?0 [ns]Bc 0.9 [MHz]Coherence Bandwidth vs. RMS Delay Spread  TX Low, Cross-RowTX High, Cross-RowTX Low, Along-RowTX High, Along-RowBc0.9=0.9/?0Bc0.9=0.63/?0  1104.5 Discussion and Implications  4.5.1 Interpretation  We found that the ?slope plus exponential? model proposed by Erceg et al. [168] fit our data well. This model was originally developed for directional antennas which imposed ?structure? on the impulse response by more strongly attenuating multipath components that arrived at the receiver from greater azimuth angles (and therefore greater path lengths). The authors observed that the model did not provide a good fit to measurements obtained with omnidirectional antennas (such as those used here) and proposed that, for such antennas, the absence of sidelobes in the antenna pattern resulted in a stronger dependence on the presence (or lack thereof) of individual scatters in the vicinity of the antennas. However, the good fit of this model to our measurements suggests that the presence of dense vegetation in the propagation path has a similar effect in more strongly attenuating echoes with longer paths.  We observed that the slope-method (i.e., .4) of estimating delay spread produced a stronger correlation with both distance and path loss than estimates obtained using the moment method (i.e., .TU). This is likely because the moment-method is an estimate of what is observed at a given signal-to-noise level, and the ability to accurately measure the underlying characteristics of the channel decreases with SNR. By contrast, the slope-method appears to be more robust at accurately measuring the ?intrinsic? delay spread due to the environment, even in low SNR conditions.  We found that the magnitude of the change in delay spread with distance and path loss (i.e., the effect size) was about 3-5 times larger (depending on the transmitting antenna height) for the cross-row case than for the along-row case. This seems to suggest that for the along-row case, paths to more distant scatterers are quickly attenuated by foliage, and only paths with similar lengths to the direct path reach the receiver with sufficient power to affect the delay spread. This may imply a ?canyon? effect whereby signals are ?guided? by the rows of trees to the reciever, similar to the effect proposed by Ghoraishi et al. which was caused by inhomogeneity in the density of forest vegetation[162].    1114.5.2 Design  In this study we observed that, for all configurations and antenna separations considered, .TU was less than 16 ns and .4 was less than 11 ns. These values are significantly lower than those typically observed for other outdoor environments (e.g., 1000?s of ns for macrocell systems at 910 MHz [172], 100?s of ns in forests at 5 GHz [167]). Systems deployed in environments similar to the one studied here, would therefore be able to tolerate faster symbol rates without incurring additional inter-symbol interference. The range of delay spreads observed here were similar to other studies conducted in forests at similar frequencies and antenna separations (e.g., [162]and[166]).  We found that the inverse relationship observed by Cox [172] and Yang [171] between .TU and coherence bandwidth at a correlation level of 0.9 fit our measurements well for a coefficient value of 0.9. The degree of agreement decreased, however, for high estimates of .TU. Excluding the cases where .TU r 7 ns (which we can see in Figure 25 correspond to cross-row propagation and ant. separation > 20m), we observed that for all measurement locations coherence bandwidth was > 10 MHz. This suggests that systems with bandwidth ? 10 MHz will not experience frequency selective fading and may be able to forego complicated equalizers, thereby increasing the energy efficiency of nodes.  4.5.3 Deployment  Path loss was observed to increase linearly with the logarithm of antenna separation for all four configurations considered in this study. Cross-row propagation at both transmitter heights was observed to experience greater path loss (and larger path loss exponents) than for along-row propagation. In the cross-row case, we observed that raising the transmitting antenna above the canopy reduced the path loss by 5-10 dB. However, for along-row propagation, raising the transmitting antenna only reduces path loss when the antennas are separated by less than 20-25 m, after which point the low antenna and high antenna configurations experience roughly the same path loss. This observation could be explained by the fact that the difference between the proportion of the direct path that is obstructed by   112vegetation for low and high transmitter height at close range is much greater than for longer range links.   The difference in path loss experienced for different propagation directions suggests that deployment scenarios in such environments need not be uniformly spaced in both directions. Wireless networks deployed here could have longer links in the along-row direction, but nodes that communicate via cross-row propagation must be more closely spaced. The results presented here further suggest that cross-row link ranges could be increased by elevating gateway or relay nodes above the canopy, though it would have little effect for along-row propagation at distances greater than 20m. It is worth noting that for typical system parameters for wireless sensor networks (TX power = 0 dBm, RX sensitivity = -100 dBm), none of the four configurations considered here would be able to reach antenna separations of 100 m.  Delay spread was observed to increase with antenna separation for cross-row propagation. Therefore the node spacing of a particular deployment will have a significant effect on the amount of multipath propagation experienced by nodes. Delay spread was also observed to decrease with increasing TX height at a given distance for cross-row propagation. Therefore, if gateway/relay nodes are required to increase the coverage area or reliability of a network deployed in this environment, these nodes will experience shorter delays.      113Chapter 5: Conclusions and Recommendations Although interest in using wireless sensor and actuator networks (WASNs) to support precision agriculture dates back nearly a decade, the field is still in its earliest stages and much work remains before deployment and configuration of such networks is reduced to standard practice. This thesis makes three principal contributions to the field, as described below, together with six recommendations for future work.  1. Assessment of the state of the art in deployment and configuration of wireless sensor and actuator networks for precision agriculture and recommendations for next steps.  This portion of our work was presented in Chapter 2 and includes surveys of: a) recent progress in developing propagation and channel models relevant to agricultural environments, b) lessons learned from the numerous demonstrations and field trials that have been conducted during the past decade, and c) techniques for enhancing wireless sensor and actuator network performance that may be applied during deployment, at the MAC layer or at the network layer. Propagation Data and Models. We found that the majority of propagation data or models relevant to wireless links that pass through vegetation, and which are currently available in the research literature, are focused on natural forest and woodland environments in which: 1) the canopies are very tall and change little in height during the course of the growing season and 2) individual trees are randomly distributed across the landscape. By contrast, agricultural environments tend to be characterized by a variety of crops that have: 1) much shorter canopies that may evolve greatly during the course of the growing season and 2) much thinner stalks and trunks that are laid out in neat rows. The lack of propagation data and models that apply to agricultural scenarios, especially those that capture the frequency, seasonal and canopy height dependence of path loss, makes it difficult for designers to conduct realistic simulations or assess the performance improvement that can be achieved through alternative deployment scenarios.  Recommendation 1: Production of propagation models useful in the design and simulation of WSANs for precision agriculture should be given high priority by the research community. Propagation researchers must work closely with system designers to ensure that those aspects   114of the propagation environment that are most relevant are identified and captured. Support for such efforts from a standards body or other association would likely be very helpful. Demonstrations and Field Trials. We found that numerous demonstrations and field trials of WASNs for precision agriculture have been conducted during the past decade and have, for the most part, proven that it is generally feasible to interconnect sensors, actuators and controllers using short-range wireless technology.  Although sensors and actuators suitable for use in agricultural environments are a mature technology and wireless communications devices suitable for use in sensor networks are widely available, adoption of wireless sensor and actuator networks for precision agriculture is proceeding at a relatively slow pace. We believe that the turning point will likely occur when decision support systems become capable of performing more effectively or reliably than a human farmer or orchardist. Until then, we believe will be difficult to achieve the full potential or justify the expense of such installations.      Recommendation 2: Development of advanced decision support systems that allow wireless sensor and actuator networks for precision agriculture to deliver their full potential should be given high priority by the research community. Network Performance Enhancement. The throughput and latency requirements of wireless sensor and actuator networks intended for use in agricultural environments are relatively modest. Reduction of energy consumption in order to ensure that battery life extends across the entire growing season is the overarching concern. We found that a number of schemes for enhancing network performance in such environments have been developed in recent years. Such schemes adopt strategies during deployment and in the design of the physical, MAC and network layers that exploit the unique aspects of wireless propagation in such environments to reduce power consumption and increase link reliability. To the best of our knowledge, however, few if any of these schemes have been routinely adopted.  Recommendation 3: Development of an industry standard for wireless sensor and actuator networks that are intended for use in agricultural environments and which incorporates best practices during deployment and in the design of the physical, MAC and network layers should be given high priority by the wireless and agricultural industry associations. During the standardization process, the performance of each proposed enhancement will be fairly   115compared and the results used to aid in the selection and incorporation of those that offer the best performance.         2. A proposal that low-power short-range wireless devices intended for use as sensor nodes in precision agriculture be allowed to share the 433 MHz sub-band currently authorized for use by active RFID devices at cargo terminals, port facilities and warehouses so that they may experience less path loss and achieve greater range and reliability while consuming less power. This portion of our work was presented in Chapter 3 and includes: a) a review of the requirements and challenges associated with deployment and configuration of wireless sensor and actuator networks in agricultural environments, b) a survey of relevant previous work including the strategies used in past efforts to secure wireless spectrum for short-range wireless devices, c) key issues in spectrum allocation, d) assessment potential bands for short-range wireless devices for precision agriculture.     We found that the majority of agricultural WSANs operate in the 915 MHz or 2.4 GHz ISM bands mainly because suitable wireless transceivers that operate in these bands are readily available. Such devices must last for the length of the growing season and must generally be battery operated as they are often deployed far from convenient sources of power. Because the nodes of such networks are often deployed within the foliage to meet operational requirements they experience relatively high path loss at these frequencies. When the lengths between nodes exceed tens of metres, closing the link can be problematic.. Although increasing receiver sensitivity or employing relay nodes may partially address these issues, a possibly simpler solution is to move to a lower carrier frequency possibly in the low UHF range. Here, based on models of path loss through vegetation presented in Rec. ITU-R P.833-7 and other works in the open literature we assess the increase in transmitter-receiver separation that can be achieved by: 1) lowering the carrier frequency, 2) increasing the receiver sensitivity. The lack of path loss models that correspond to specific agricultural use cases of interest necessitates the use of models that were developed for forests and woodlands and therefore can only provide a rough estimate of the improvements one can anticipate by using the above methods.    116Recommendation 4: The potential role of relays and gateways to allow sensor nodes to be more widely separated should be investigated. This will require more sophisticated path loss models than are currently available in the research literature. In particular, models that capture the path loss experienced between nodes located above and within the canopy in the along-row and cross-row directions of a typical agricultural field must be available.    By comparing these estimates with the requirements of agricultural links, we assess the suitability of available ISM bands and potential alternatives. Our results suggest that the 433 MHz sub portion of the 43? ? 438 MHz band is an attractive alternative to the 915 and 2450 MHz ISM bands and that steps should be taken to globally harmonize its use by short-range wireless devices in agricultural environments.  Recommendation 5: A dialog between academic researchers, agricultural and wireless industry leaders, and spectrum regulators including relevant ITU-R study groups concerning the potential role for a low-UHF spectrum allocation for WASNs in agriculture should be conducted. ITU-R and national spectrum regulators will not consider requests for spectrum allocation without significant interest and a strong business case from industry. Such a discussion should take place before the industry invests too heavily in high UHF devices.        1173. Interpretation and Implications of the Channel Impulse Response Observed at 2450 MHz in High Density Apple Orchards This portion of our work was presented in Chapter 4 and includes surveys of: a) recent progress in developing propagation and channel models relevant to agricultural environments, b) lessons learned from the numerous demonstrations and field trials that have been conducted during the past decade, and c) techniques for enhancing wireless sensor and actuator network performance that may be applied during deployment, at the MAC layer or at the network layer. Interest in deploying WSAN to support precision agriculture in both farm and orchard environments, including high-density apple orchards consist of long rows of closely spaced, vertically oriented branches planted directly in the ground, has increased dramatically in recent years. Although the nodes of such networks are often deployed within the foliage in order to meet operational requirements, gateway and relay nodes may be deployed above or within the foliage in order to achieve the required coverage and connectivity.   The manner in which a wireless channel within such a network temporally disperses a signal may depend greatly on the deployment scenario, can be used to identify dominant propagation paths and components within the scenario, and affects key aspects of physical layer design. The majority of past efforts to characterize channel impulse response when one or both ends of the wireless link are deployed within foliage have focused on natural forests and woodlands that are characterized by a uniform distribution of trunks under a homogenous canopy layer.  However, the results of such efforts may not apply to high-density apple orchards which have virtually no trunk-layer and which present very different geometries in the along-row and cross-row propagation directions. Here, we present measurements and analysis of the channel impulse response of a high-density apple orchard between 2 and 3 GHz. We reveal that, in most cases, the delay spread in this environment is sufficiently short (or, equivalently, the frequency response is sufficiently flat) to simplify or eliminate the need for complicated equalizers, thereby reducing the power consumption of nodes. Further, we reveal that delay spread in this environment is significantly shorter than for conventional indoor environments, and therefore systems operating in such orchards will experienced less inter-symbol interference.    118We also reveal that values of delay spread experienced in this environment are similar to those experienced in natural forests for similar frequencies and antenna separations. We also demonstrate that delay spread is shorter for propagation along rows than for across rows that may be due to a ?canyon? effect whereby the rows of trees ?guide? reflected signals towards the receiver. Cross-row propagation would therefore lacks this characteristic and signals would instead be reflected from more distant scatterers. We reveal that delay spread decreases when the transmitter is elevated above the tree canopy, though only for cross-row propagation and that delay spread increases linearly with antenna separation. Accordingly, designers must consider the delay spread experienced at maximum range when developing systems to operate in these environments. Finally, we show that there is a strong positive correlation between path loss and delay spread, though the size of the effect is much larger for cross-row propagation than for along-row propagation. Recommendation 6: While our data set represents a useful beginning, further measurement data should be collected across a variety of crops and locations and a more comprehensive joint path-loss-delay-spread model created. Such a model will help designers more accurately assess the performance that can be achieved using different network layouts and configurations. Such an effort should be coordinated, and at the very least, endorsed, by a major wireless or agricultural industry association.    119 References [1] M. A. Oliver, "Geostatistics and precision agriculture: A way forward," in Geostastical Applications for Precision Agriculture. Netherlands: Springer, 2010. [2] J. K. Schueller, "A review and integrating analysis of spatially-variable control of crop production," Fertilizer Research, vol. 33, pp. 1-34, 1992. [3] N. Zhang, M. Wang, and N. Wang, "Precision agriculture - A worldwide overview," Computers and Electronics in Agriculture, vol. 36, no. 2-3, pp. 113-132, November 2002. [4] S. Cox, "Information technology: the global key to precision agriculture and sustainability," Computer Electronics in Agriculture, vol. 36, no. 2-3, pp. 93-111, November 2002. [5] A. McBratney, B. Whelan, and T. Ancev, "Future directions of precision agriculture," Precision Agriculture, vol. 6, no. 1, pp. 7-23, 2005. [6] N. Wang, N. Q. Zhang, and M. H. Wang, "Wireless sensors in agriculture and food industry - Recent development and future perspective," Computer Electronics in Agriculture, vol. 50, no. 1, pp. 1-14, January 2006. [7] L. Ruiz-Garcia, L. Lunadei, P. Barreiro, and J. I. Robla, "A review of wireless sensor technologies and applications in agriculture and food industry: State of the art and current trends," Sensors, vol. 9, pp. 4728-4750, 2009. [8] R. Gebbers and V. I. Adamchuck, "Precision agriculture and food security," Science, vol. 327, no. 5967, pp. 828-831, February 2010. [9] A. Rheman, A. Z. Abbasi, N. Islam, and Z. A. Shaikh, "A review of wireless sensors and networks' applications in agriculture," Computer Standards and Interfaces. [Online]. http://dx.doi.org/10.1016/j.csi.2011.03.004 [August 25, 2013]. [10] A. Mainwaring, J. Polastre, R. Szewczyk, D. Culler, and J. Anderson, "Wireless sensor networks for habitat monitoring," in Proceedings of Wireless Sensor Networks and Applications '02, Atlanta, GA, 2002. [11] W. Chebbi, M. Benjemaa, A. Kamoun, M. Jabloun, and A. Sahli, "Development of a WSN integrated weather station node for an irrigation alert program under Tunisian conditions," in Proceedings of Internation Conference on Systems, Signals, and Devices, 2011, pp. 1-6. [12] F. J. Pierce and T. V. Elliott, "Regional and on-farm wireless sensor networks for agricultural systems in Eastern Washington," Computer Electronics in Agriculture, vol. 61, no. 1, pp. 32-43, April 2008. [13] T. Arici and Y. Altunbasak, "Adaptive sensing for environment monitoring using wireless sensor networks," in Proceedings of Wireless Communications and Networking Conference, 2004, pp. 2347-2352. [14] I. F. Akyildiz, "Wireless sensor networks: A survey," Computational Networks, vol. 38, no. 4, pp. 393-422, March 2002. [15] P. Andrade-Sanchez, F. J. Pierce, and T. V. Elliot, "Performance assessment of wireless sensor netowrks in agricultural settings," ASABE, Paper No. 073076, 2007. [16] H. Charles et al., "Food security: The challenge of feeding 9 billion people," Science,   120vol. 327, no. 5967, pp. 812-818, February 2010. [17] D. Tilman, K. G. Cassman, P. A. Matson, R. Naylor, and S. Polasky, "Agricultural sustainability and intensive production practices," Nature, vol. 418, pp. 671-677, August 2002. [18] M. Qadir, T. M. Boers, S. Schubert, A. Ghafoor, and G. Murtaza, "Agricultural water management in water-starved countries: challenges and opportunities," Agricultural Water Management, vol. 62, no. 3, pp. 165-185, October 2003. [19] P. Gruhn, F. Goletti, and M. Yudelman, "Integrated nutrient management, soil fertility, and sustainable agriculture: Current issues and future challanges," International Food Policy Research Institute, Washington, D.C., Food, Agriculture, and the Environment Discussion Paper no. 32 2002. [20] "Invasive alien species: No rest from new pests," BC Ministry of Agriculture and Lands, 2007. [Online]. http://www.agf.gov.bc.ca/cropprot/ias.pdf [August 25, 2013]. [21] D. M. Huber et al., "Invasive pest species: Impacts on agricultural production, natural resources, and the environment," Council for Agricultural Science and Technology (CAST), Issue Paper no. 20, 2002. [22] R. N. Mack et al., "Biotic invasions: Causes, epidemiology, global consequences and control," Ecological Applications, vol. 10, no. 3, pp. 689-710, 2000. [23] K. G. Cassman, "Ecological intensificationof cereal production systems: Yield potential, soil quality, and precision agriculture," Proceedings of the National Academy of Science USA, vol. 96, pp. 5952-5959, May 1999. [24] M. A. Oliver, "An overview of geostatistics and precision agriculture," in Geostastical Applications for Precision Agriculture. Netherlands: Springer, 2010. [25] A. Dobermann, S. Blackmore, S. E. Cook, and V. I. Adamchuck, "Precision farming: challenges and future directions," in Proceedings of 4th International Crop Science Congress, Brisbane, Australia, 2004. [26] A. O. Ani, F. O. Okwudiuche, C. C. Anyadike, N. Onuoha, and B. B. Uzoejinwa, "Applying the concepts of precision agriculture to tillage," in Proceedings of the International Soil Tillage Research Organization Nigeria Symposium on Tillage for Agricultural Productivity and Envrionmental Sustainability, 2011, pp. 241-251. [27] U.S. Geological Survey. (2011, April) Landsat 1 History. [Online]. http://landsat.usgs.gov/about_landsat1.php [August 25, 2013]. [28] J. R. Jensen, J. E. Estes, L. W. Bowden, and L. R. Tinney, "Remote sensing of water demand information," Geographical Review, vol. 67, no. 3, pp. 322-334, July 1977. [29] G. Konecny, "Mapping from space," Remote Sensing For Enironmental Data in Albania: A Strategy for Integrated Management, NATO Science Series, vol. 72, pp. 41-58, 2000. [30] S. Moran. (2012, May) Precision farming land managment. [Online]. http://landsat.gsfc.nasa.gov/about/Application1.3.html [August 25, 2013]. [31] D. L. Williams, "Computer analysis and mapping of gypsy moth defoliation levels in Pennsylvania using Landsat-1 digital data," Proceedings of the NASA Earth Presentations, vol. 1A, pp. 167-181, 1975. [32] T. Brase. Online companion: Precision Agriculture. [Online].   121http://www.delmarlearning.com/companions/index.asp?isbn=140188105X [August 25, 2013]. [33] V. I. Adamchuck and R. B. Ferguson. (2009) Site Specific Crop Managment. [Online]. http://bse.unl.edu/adumchuk/class_ssm/2009/2009.html [August 25, 2013]. [34] P. C. Robert, "Precision agriculture: A challange for crop nutrition management," Plant Soil, vol. 247, pp. 143-149, 2002. [35] D. C. Jochinke, B. J. Noonon, N. G. Wachsmann, and R. M. Norton, "The adoption of precision agriculture in Australian broadacre cropping system - Challenges and opportunites," Field Crop Research, vol. 104, no. 1-3, pp. 68-87, December 2007. [36] D. L. Ndzi et al., "Vegetation attenuation measurements and modeling in plantations for wireless sensor network planning," Progress in Electromagnetics Research B, vol. 36, pp. 283-301, 2012. [37] P. Mestre, J. Ribeiro, C. Serodio, and J. Monteiro, "Propagation of IEEE802.15.4 in vegetation," in Proceedings of the World Congress on Engineering , London, U.K., 2011. [38] S. Phaiboon and S. Somkuarnpanit, "Mobile path loss characteristics for low base station antenna height in different forest densities," in 1st International Symposium on Wireless Pervasive Computing, 2006. [39] J. A. Gay-Fernandez, M. G. Sanchez, I. Cuinas, and A. V. Alejos, "Propagation analysis and deployment of a wireless sensor network in a forest," Progress in Electromagnetics Research, vol. 106, pp. 121-145, 2010. [40] I. J. Dilworth and B. L'Ebraly, "Propagation effects due to foliage and building scatter at millimetre wavelengths," in Proceedings of 9th IEEE International Conference on Antennas and Propagation, Eindhoven, Netherlands, 1995, pp. 51-53. [41] J. E. J. Dalley and M. S. Smith, "Propagation losses due to foliage at various frequencies," in Proceedings of IEE National Conference on Antennas and Propagation, York, U.K., 1999, pp. 267-270. [42] S. Vougioukas, H. T. Anastassiu, C. Regen, and M. Zude, "Influence of foliage on radio path losses (PLs) for wireless sensor network (WSN) planning in orchards," Biosystems Engineering, pp. 1-12, 2012. [43] A. Harun et al., "Comparative performance analysis of wireless RSSI in wireless sensor networks motes in tropical mixed-crop precision farm," in 3rd International Conference on Intelligent Systems Modelling and Simulation, 2012, pp. 606-610. [44] J. Thelen, D. Goense, and K. Langendoen, "Radio wave propagation in potato fields," in 1st Workshop on Wireless Network Measurements, Riva del Garda, Italy, 2005. [45] H. Liu, Z. Meng, and Y. Shang, "Sensor nodes placement for farmland enviornmental monitoring applications," in WiCom '09, 2009, pp. 1-4. [46] S. Li and H. Gao, "Propagation characteristics of 2.4 GHz wireless channel in cornfields," in Proceedings of International Conference on Communications Technology, 2011, pp. 136-140. [47] A. R. Silva and M. C. Vuran, "Empirical evaluation of wireless underground-to-underground communication in wireless underground sensor networks," Lecture Notes in Computing Science, vol. 5516, pp. 231-244, 2009.   122[48] X. Yu, P. Wua, W. Hana, and Z. Zhang, "A survey on wireless sesnor networks infrastructure for agriculture," Computer Standards and Interfaces, vol. 35, no. 1, pp. 59-64, January 2013. [49] H. St. Michael and I. Otung, "Characterization and prediction of excess attenuation of microwave radio signals by vegetation forms," in Proceedings of 12th IEE International Conference on Antennas and Propagation, Exeter, U.K., 2003, pp. 637-640. [50] N. Savage, D. Ndzi, A. Seville, E. Vilar, and J. Austin, "Radio wave propagation through vegetation: Factors influencing signal attenuation," Radio Science, vol. 38, no. 5, p. 1088, 2003. [51] K. Sarabandi and I. S. Koh, "Effect of canopy-air interface roughness on HF-VHF wave propagation in forest," IEEE Transactions on Antennas and Propagation, vol. 50, no. 2, pp. 111-121, 2002. [52] E. R. Pelet, J. E. Salt, and G. Wells, "Effect of wind on foliage obstructed line-of-sight channel at 2.5 GHz," IEEE Transactions on Broadcasting, vol. 50, no. 3, pp. 224-232, 2004. [53] G. Acquaah, Principles of Crop Production: Theory, Techniques, and Technology. Upper Saddle River, N.J.: Pearson Prentice Hall, 2005. [54] K. M. Moncada and C. C. Shaeffer, Introduction to Agronomy: Food, Crops, and Environment. Clifton Park, N.Y.: Delmar Cengage Learning, 2009. [55] N. S. Goel and D. E. Strebel, "Simple beta distribution representation of leaf orientation in vegetation canopies," Agronomy Journal, vol. 76, no. 5, pp. 503-513, 2004. [56] J-P. P. Wigneron et al., "Characterizing the dependence of vegetation model parameters on crop structure, incidence angle, and polarization at L-band," IEEE Transactions on Geoscience and Remote Sensing, vol. 42, no. 2, pp. 416-425, 2004. [57] F. Marliani, "Simulating coherent backscattering from crops during the growing cycle," IEEE Transactions on Geoscience and Remote Sensing, vol. 40, no. 1, pp. 162-177, 2002. [58] X. Zuo et al., "An environment monitoring system for valuable chinese herbal medicine growing based on wireless sensor networks," in Proceedings of World Automation Congress, 2010, pp. 71-75. [59] Z. Feng, "Research on water-saving irrigation automatic control system based on internet of things," in Proceedings of International Conference on Electronics, Information and Communication Engineering, 2011, pp. 2541-2544. [60] M. Mancuso and F. Bustaffa, "A wireless sensors network for monitoring environmental variables in a tomato greenhouse," in Proceedings of IEEE International Workshop on Factory Communication Systems, 2009, pp. 107-110. [61] D. Anurag, S. Roy, and S. Bandyopadhyay, "Agro-Sense: Precision agriculture using sensor-based wireless mesh networks," in Proceedings of Innovations in NGN: Future Network and Services. 1st ITU-T Kaleidoscope Academic Conference, 2008, pp. 383-388. [62] L. L. Li, S. F. Yang, L. Y. Wang, and X. M. Gao, "The greenhouse environment   123monitoring system based on wireless sensor network technology," in Proceedings IEEE International Conference on Cyber Technology in Automation, Control, and Intelligent Systems, 2011, pp. 265-268. [63] D. Grant and A. Al-Anbuky, "Wireless microclimate sensor," in Proceedings of International Conference on Intelligent Sensors, Sensor Networks, and Information Processing, 2007, pp. 663-668. [64] A. Singh, L. S. Chyan, and P. Sebastian, "Sensor integration in wireless sensor network system for environmental monitoring system," in Proceedings of International Conference on Intelligent and Automation Systems, 2010, pp. 1-5. [65] G. Vellidis, M. Tucker, C. Perry, C. Kvien, and C. Bednarz, "A real-time wireless smart sensor array for scheduling irrigation," Computers and Electronics in Agriculture, vol. 61, no. 1, pp. 44-50, April 2008. [66] S. Verma, N. Chug, and V. Gadre, "Wireless sensor network for crop field monitoring," in Proceedings of International Conference on Recent Trends in Information, Telecommunication and Computing, 2010, pp. 207-211. [67] Y. Jiber, H. Harroud, and A. Karmouch, "Precision agriculture monitoring framework based on WSN," in Proceedings of International Wireless Communications and Mobile Computing Conference, 2011, pp. 2015-2020. [68] A. Baggio, "Wireless sensor networks in precision agriculture," in Preceedings of Workshop on Real-World Wireless Sensor Networks, Stockholm, 2005. [69] K. Langendoen, A. Baggio, and O. Visser, "Murphy loves potatoes: Experiences from a pilot sensor network deployment in precision agriculture," in Proceedings of Parallel and Distributed Processing Symposium, 2006. [70] R. Beckwith, D. Teibel, and P. Bowen, "Unwired wine: sensor networks in vineyards," Proceedings of IEEE Sensors, vol. 2, pp. 561-564, 2004. [71] H. Liu, Z. Meng, and S. Cui, "A wireless sensor network prototype for environmental monitoring in greenhouses," in Proceedings of International Conference on Wireless Communications, Networking and Mobile Computing, 2007, pp. 2344-2347. [72] E. S. Nadimi, H. T. Sogaard, T. Bak, and F. W. Oudshoorn, "ZigBee-based wireless sensor networks for monitoring animal presence and pasture time in a strip of new grass," Computer Electronics in Agriculture, vol. 61, pp. 79-87, 2008. [73] G. Anastasi, O. Farruggia, G. L. Re, and M. Ortolani, "Monitoring high-quality wine production using wireless sensor networks," in Proceedings of 42nd Hawaii International Conference on System Sciences, 2009, pp. 1-7. [74] K. H. Kwong et al., "Adaptation of wireless sensor network for farming industries," in Proceedings of 6th International Conference on Networked Sensing Systems, 2009, pp. 1-4. [75] L. Bencini et al., "Agricultural monitorning based on wireless sensor network technology: Real long life deployments for physiology and pathogens control," in Proceedings 3rd International Conference on Sensor Technologies and Applications, Athens/Glyfada, Greece, 2009, pp. 372-377. [76] S. Li, J. Cui, and Z. Li, "Wireless sensor network for precise agriculture monitoring," in Proceedings of International Conference on Intelligent Computation Technology   124and Automation, 2011, pp. 307-310. [77] V. Jelicic, T. Razov, D. Oletic, M. Kuri, and V. Bilas, "MasliNET: A wireless sensor network based environmental monitoring system," in Proceedings of International Convention on Information and Communication Technology, Electronics and Microelectronics, 2011, pp. 150-155. [78] F. J. Pierce et al., "A remote-real-time continuous move irrigation control and monitoring system," American Society of Agricultural and Biological Engineers, St. Joseph, MI, Paper No. 062162 2006. [79] M. Martinelli, L. Ioriatti, F. Viani, M. Benedetti, and A. Massa, "A WSN-based solution for precision farm purposes," in Proceedigns of IEEE Geoscience and Remote Sensing Symposium, Cape Town, South Africa, 2009. [80] Y. Kim, R. G. Evans, and W. M. Iverson, "Remote sensing and control of an irrigation system using a distributed wireless sensor network," IEEE Transactions on Instrumentation and Measurment, vol. 57, no. 7, pp. 1379-1387, July 2008. [81] L. Shengduo, "A water-saving irrigation monitoring system based on wireless data acquisition network," Software Engineering and Knowledge Engineering, AISC, vol. 114, pp. 773-780, 2012. [82] J. Panchard, S. Rao, T. V. Prabhakar, H. S. Jamadagni, and J. P. Hubaux, "Common-sense net: Improved water management for resource-poor farmers via sensor networks," in Proceedings of International Conference on Information and Communications Technologies and Development, 2006, pp. 22-33. [83] C. J. Ritsema et al., "A new wireless underground network system for continuous monitoring of soil water contents," Water Resources Research, vol. 45, no. 4, pp. 1-9, 2009. [84] P. Marino, F. P. Fontan, M. A. Dominquez, and S. Otero, "Internetworking infrastructures for field sensors," in Proceedings of International Workshop on Factory Communication Systems, 2008, pp. 113-116. [85] Q. Ren and C. Chang, "A grain storage monitoring system based on wireless sensor networks," in Proceedings of International Conference on Communications Technology and Applications, 2009, pp. 308-311. [86] O. Green et al., "Montoring and modeling temperature variations inside silage stacks using novel wireless sensor networks," Computers and Electronics in Agriculture, vol. 69, no. 2, pp. 149-157, December 2009. [87] K. H. Kwong et al., "Implementation of a herd managment system with wireless sensor networks," IET Wireless Sensor Systems, vol. 1, no. 2, pp. 55-65, 2011. [88] K. Mayer, K. Taylor, and K. Ellis, "Cattle health monitoring using wireless sensor networks," in Proceedings of 2nd IASTED International Conference on Communication and Computer Networks, Cambridge, Massachusetts, 2004. [89] J. Hwang, C. Shin, and H. Yoe, "Study on an agricultural environment monitoring server system using wireless sensor networks," Sensors, vol. 10, pp. 11189-11211, 2010. [90] G. Zhu, "The application of wireless sensor networks in management of orchard," in Computer and Computing Technologies in Agriculture III, Berlin: Springer, 2010.   125[91] L. Zhaohua and S. Meijun, "A long-range wireless greenhouse monitoring system using wireless sensor networks," in Proceedings of 2nd International Conference on Industrial and Information Systems, 2010, pp. 35-39. [92] L. Yumei, Z. Changli, and Z. Ping, "The temperature humidity monitoring system of soil based on wireless sensor networks," in Proceedings of International Conference on Electronics, Information and Communication Engineering, 2011, pp. 1850-1853. [93] W. Zhang, G. Kantor, and S. Sing, "Integrated wireless sensor/actuator networks in an agricultural application," in Proceedings ACM Congress on Embedded Networked Sensor Systems, 2004, p. 317. [94] J. A. L. Riquelme et al., "Wireless sensor networks for precision horticulture in Sourthern Spain," Computers and Electronics in Agriculture, vol. 68, no. 1, pp. 25-35, August 2009. [95] J. M. Tarara, J. C. Ferguson, P. E. Blom, M. J. Pitts, and F. J. Pierce, "Automated estimation of grapevine yields via trellis tension," Transactions of ASAE, vol. 47, no. 2, pp. 647-657, 2004. [96] I. F. Akyildiz, S. Weilian, Y. Sankarasubramaniam, and E. Cayirci, "A survey on sensor networks," IEEE Communications Magazine, vol. 40, no. 8, pp. 102-114, August 2002. [97] G. T. Flatman and A. A. Yfantis, "Geostatistical stategy for soil sampling: The survey and the census," Environmental Monitoring Assessment, vol. 4, no. 4, pp. 335-349, 1984. [98] M. Keshtgari and A. Deljoo, "A wireless sensor network solution for precision agriculture technology," Wireless Sensor Networks, vol. 4, no. 1, pp. 25-30, 2012. [99] K. Konstantinos, X. Apostolos, K. Panagiotis, and S. George, "Topology optimization in wireless sensor networks for precision agriculture applications," in Proceedings of International Conference on Sensor Technologies and Applications, 2007, pp. 526-530. [100] B. Hu et al., "Impact of vector quantization compression on hyperspectral data in the retrieval accuracies of crop chlorophyll content for precision agricutlure," in Proceedings of IEEE Geoscience and Remote Sensing Symposium, 2002, pp. 1655-1657. [101] K. P. Ferentinos, T. A. Tsiligiridis, and K. G. Arvanitis, "Energy optimization of wireless sensor networks for environmental measurements," in Proceedings of IEEE International Conference on Computational Intelligance for Measurement Systems and Applications, 2005, pp. 250-255. [102] H. Sahota, R. Kumar, A. Kamal, and J. Huang, "An energy-efficient wireless sensor network for precision agriculture," in Proceedings IEEE Symposium on Computers and Communications, 2010, pp. 347-350. [103] A. El-Hoiydi and J. D. Decotignie, "WiseMAC: an ultra low power MAC protocol for the downlink of intrastructure wireless sensor networks," in Proceedings of 9th International Symposium on Computers and Communications, 2004, pp. 244-254. [104] F. Chiti et al., "Design and application of enhanced communication protocols for wireless sensor networks operating in environmental monitoring," in Proceedings of IEEE Conference on Communications, vol. 8, 2006, pp. 3390-3395.   126[105] K. Langdoen and A. Meier, "Analyzing MAC protocols for low data-rate application," ACM Transactions on Sensor Networks, vol. 7, no. 2, p. 19, 2010. [106] F. El-Moukaddem, E. Torng, G. Xing, and S. Kulkarni, "Mobile relay configuration in data-intensive wireless sensor networks," in Proceedings of IEEE 6th International Conference on Mobile Adhoc and Sensor Systems, 2009, pp. 80-89. [107] N. P. Karthickraja, V. Sumathy, and J. Ahamed, "A novel hybrid routing protocol for data aggregation in agricultural applications," in Proceedings IEEE International Conference on Communication Control and Computing Technologies, 2010, pp. 227-231. [108] R. Aquino-Santos, A. Gonzalez-Potes, A. Edwards-Block, and R. A. Virgen-Ortiz, "Developing a new wireless sensor network platform and its application in precision agriculture," Sensors, pp. 1192-1211, 2011. [109] S. Sutar, S. Jayesh, and K. Priyanka, "Irrigation and fertilizer control for precision agriculture using WSN: Energy efficient approach," International Journal of Advances in Computing and Information Researches, vol. 1, no. 1, 2012. [110] N. Pradhan and T. Saadawi, "Impact of physical propagation environment and traffic load on the performance of routing protocols ," in Proceedings of International Conference on Innovations in Information Technology, 2008, pp. 529-533. [111] A. Rhattoy and A. Zatni, "The impact of propagation environment and traffic load on the performance of routing protocols in ad hoc networks," International Journal of Distributed and Parralel Systems, vol. 3, no. 1, pp. 75-87, January 2012. [112] ITU-R Recommendation M.2002, "Objectives, characteristics and functional requirements of wide-area sensor and/or actuator network (WASN) systems," 2012. [113] "Framework for the use of certain non-broadcasting applications in the television broadcasting bands below 698 MHz," Industry Canada, IC SMSE-012-12 2012. [114] "Promoting the shared use of radio spectrum resources in the internal market," European Commission, Brussels, COM 478 Final, 2012. [115] J. Alden, "Exploring the value and economic valuation of spectrum ," ITU, 2012. [116] Z. L. Frogbrook, M. A. Oliver, M. Salahi, and R. H. Ellis, "Exploring the spatial relations betweeen cereal yield and soil chemical properties and the implications for sampling," Soil Use and Management, vol. 18, pp. 1-9, 2002. [117] J. Yana, C. K. Lee, M. Umeda, and T. Kosaki, "Spatial variability of soil chemical properties in a paddy field ," Soil Science and Plant Nutrition, vol. 46, no. 2, pp. 473-482, 2000. [118] Q. Jiang, Q. Fu, and Z. Wang, "Delineating site-specific irrigation management zones," Irrigation and Drainage, vol. 60, pp. 464-472, 2011. [119] G. Yang and M. Xiao, "Node placement strategies to prolong lifetime in one-dimensional linear wireless sensor network," Applied Mechanics and Materials, vol. 55-57, pp. 1705-1710, 2011. [120] P. Padilla et al., "On the influence of the propagation channel in the performance of energy-efficient geographic routing algorithms for wireless sensor networks (WSN)," Wireless Personal Communications, May 2012. [121] "Review of Part 15 and Other Parts of the Commission's Rules," Federal   127Communications Commission, Washington, D.C., FCC 04-98 2004. [122] "FCC Allocates Spectrum in 4.9 GHz Ranges for Intelligent Transportation Systems Uses," FCC, Report No. ET 99-5, 1999. [123] "Amendment of Parts 2 and 90 of the Commission's Rules to Allocate the 5.850-5.925 GHz Band to the Mobile Service for Dedicated Short Range Communications of Intelligent Transportation Services," FCC, ET Docket No. 98-95, RM-9096, 1998. [124] "Harmonisation of Frequency Bands to be Designated for Road Transport Information Systems," European Radiocommunications Committee, Lisbon, ERC Report 3, 1991. [125] "Technical Characteristics for Pan-European Harmonized Communications Equipment Operating in the 5 GHz Frequency Range and Intended for Critical Road-Safety Applications: Part 1," ETSI, TR 102492-1 2005. [126] " Technical Characteristics for Pan-European Harmonized Communications Equipment Operating in the 5 GHz Frequency Range and Intended for Critical Road-Safety Applications: Part 2," ETSI, TR 102 492-2 2005. [127] "Compatibility Studies in the Band 5855-5925 MHz Between Intelligent Transport Systems (ITS) and Other Systems," European Communications Committee, Bern, ECC Report 101, 2007. [128] "Technical Feasibility of Harmonising a Sub-band of Band IV and V for Fixed/Mobile Applications (Including Uplinks), Minimising the Impact on GE06," Electronic Communications Committee, CEPT Report 22, 2007. [129] S. Hanna, "Regulations and standards for wireless medical applications," in Proceedings of the 3rd International Symposium on Medical Information and Communication Technology, 2009. [130] ITU-R Recommendation M.2224, "System design guidelines for wide area sensor and/or actuator network (WASN) systems," 2011. [131] A. Harney and C. O'Mahony. (2007, Jul) Understand wireless short-range devices for global license-free systems. Online. [Online]. http://www.eetimes.com/document.asp?doc_id=1276268 [August 25,2013]. [132] ITU-R Recommendation P.1406-1, "Propagation effects relating to terrestrial land mobile and broadcasting services in the VHF and UHF bands," 2007. [133] ITU-R Recommendation P.1812-2, "A path-specific propagation prediction method for point-to-area terrestrial services in the VHF and UHF bands," 2012. [134] ITU-R Recommendation P.372-10, "Radio Noise," 2009. [135] ITU-R M.1079-2, "Performance and quality of service requirements for International Mobile Telecommunications-2000 (IMT-2000) access networks," 2003. [136] ITU-R SM.2153-1, "Technical and operating parameters and spectrum requirements for short range radiocommunication devices," 2012. [137] M. O. Al-Nuaimi and A. M. Hammoudeh, "Measurements and predictions of attenuation and scatter of microwave signals by trees," IEEE Proceedings of Microwaves Antennas Propagation, vol. 141, no. 2, pp. 70-76, 1994. [138] M. A. Weissberger, "An initial critical summary of models for predicting the attenuation of radio waves by foliage," Electromagnetic Compatibility Analysis Center, Annapolis, MD, ESD-TR-81-101 1981.   128[139] CCIR, "Influences of terrain irregularities and vegetation on troposphere propagation," Geneva, CCIR Report 1986. [140] Y. S. Meng, Y. H. Lee, and B. C. Ng, "Path loss modeling for near-ground VHF radio-wave propagation through forest with tree-canopy reflection effect," Progress in Electromagnetics Research M, vol. 12, pp. 131-141, 2010. [141] M. O. Al-Nuami and R. B. L. Stephens, "Measurments and prediction model optimization for signal attenuation in vegetation media at centimetre wave frequencies," IEE Proceedings of Microwaves Antennas and Propagation, vol. 145, no. 3, pp. 201-206, 1998. [142] COST 235, "Radiowave propagation effects on next generation fixed services terrestrial telecommunications systems," Commission of the European Union, Final Report ISBN 92-827-8023-6 1996. [143] N. C. Rogers et al., "A generic model of 1-60 GHz radio propagation through vegetation," Radiocommunications Agency, Technical Report 2002. [144] ITU-R Recommendation P.833-7, "Attenuation in vegetation," 2012. [145] F. K. Schewering, E. J. Violelte, and R. H. Espeland, "Millimetre-wave propagation in vegetation: Experiments and theory," IEEE Transactions Geoscience and Remote Sensing, vol. 26, no. 3, pp. 355-367, 1988. [146] A. Seville, "Vegetation attenuation: Modeling and measurments at millimetric frrequencies," in Proceedings of 10th IEE International Conference on Antennas and Propagation, 1997, pp. 5-8. [147] Y. S. Meng and Y. H. Lee, "Investigations of foliage effect on modern wireless communication systems: a review," Progress in Electromagnetics Research, vol. 105, pp. 313-332, 2010. [148] R. B. Stephens and M. O. Al-Nuaimi, "Attenuation measurement and modelling in vegetation media at 11.2 and 20 GHz," Electronics Letters, vol. 31, no. 20, pp. 1783-1785, 1995. [149] ITU RSM/07, "Background paper: Radio spectrum management for a converging world," 2004. [150] A. Seville and K. H. Craig, "Semi-empirical model for millimetre-wave vegetation attenuation rates," Electronics Letters, vol. 31, no. 17, pp. 1507-1508, 1995. [151] S. Yin, D. Chen, Q. Zhang, M. Liu, and S. Li, "Mining spectrum usage data: A large-scale spectrum measurement study," IEEE Transactions on Mobile Computing, vol. 11, no. 6, pp. 1033-1046, June 2012. [152] FCC Title 47 Part 15 Section 15.240, "Operation in the band 433.5-434.5 MHz," 2005. [153] Industry Canada RSS 210, "License-exempt radio apparatus (all frequency bands): Category I Equipment," 2010. [154] "Radiocommunications (low interference potential devices) class license 2000," ComLaw, F2013C00396, 2013. [155] "Radiocommunications regulations (general user radio licence for short range devices) notice no.2," New Zealand Gazette, no. 61, pp. 1751-1754, May 2012. [156] "Table of radio frequency allocations of the Republic of China," 2005.   129[157] C. Swedberg, "China endorses ISO 18000-7 433 MHz standard," RFID Journal, Nov. 2006. [Online]. http://www.rfidjournal.com/articles/view?2807 [August 25, 2013]. [158] "Telemeter, Telecontrol and Data Transmission Radio Equipment for Specified Low-Power Radio Station v1.1," Association of Radio Industries and Businesses, ARIB STD-T67 2005. [159] J. Liang and Q. Liang, "Outdoor propagation channel modeling in foliage enviornment," IEEE Transactions on Vehicular Technology, vol. 59, no. 5, pp. 2243-2252, June 2010. [160] R. J. C. Bultitude, "Measured characteristics of 800/900 MHz fading radio channels with high angle propagation through moderately dense foliage," IEEE Journal of Selected Areas in Communications, vol. 5, no. 2, pp. 116-127, 1987. [161] G. G. Joshi et al., "Near-ground channel measurments over line-of-sight and forested paths," IEE Proceedings Microwaves Antennas and Propagation, vol. 152, no. 6, pp. 589-596, 2005. [162] M. Ghoraishi, J. I. Takada, C. Phakasoum, and T. Imai, "Radio wave propagation through vegetation," in Wave Propagation Theories and Aplications., ch. 6. [Online]. http://www.intechopen.com/books/wave-propagation-theories-and-applications [August 25, 2013]. [163] C. Oestges, B. M. Villacieros, and D. Vanhoenacker-Janvier, "Radio channel characterization for moderate antenna heights in forest areas," IEEE Transactions on Vehicular Technology, vol. 58, no. 8, pp. 4031-4035, October 2009. [164] Y. S. Meng, Y. H. Lee, and B. C. Ng, "Investigation of rainfall effects on forested radio wave propagation," IEEE Antennas and Wireless Propagation Letters, vol. 7, pp. 159-162, 2008. [165] Y. S. Meng, Y. H. Lee, and B. C. Ng., "Further study of rainfall effect on VHF forested radio-wave propagation with four-layered model," Progress in Electromagnetics Research, vol. 99, pp. 149-161, 2009. [166] M. Ghoraishi, J. I. Takada, C. Phakasoum, T. Imai, and K. Kitao, "Azimuth and delay dispersion of mobile radio wave propagation through vegetation," in Proceedings of 4th European Conference on Antennas and Propagation, April 2010, pp. 1-4. [167] D. W. Matolak, F-C. Yang, and H. B. Riley, "Short range forest channel modeling in the 5 GHz band," in Proceedings of the 6th European Conference on Antennas and Propagation, 2012, pp. 3337-3341. [168] V. Erceg et al., "A model for the multipath delay profile of fixed wireless channels," IEEE Journal on Selected Areas in Communications, vol. 17, no. 3, pp. 399-407, March 1999. [169] M. S. Varela and M. G. Sanchez, "RMS delay and coherence bandwidth measurements in indoor radio channels in the UHF band," IEEE Transactions on Vehicular Technology, vol. 50, no. 2, pp. 515-525, March 2001. [170] H. Arslan and T. Yucek, "Estimation of frequency selectivity for OFDM based new generation wireless communication systems," in Proceedings of the World Wireless Congress, 2003. [171] H. Yang, P. F. M. Smulders, and M. H. A. J. Herben, "Frequency selectivity of 60-GHz   130LOS and NLOS indoor radio channels," in Proceedings of IEEE Vehicular Technology Conference, 2006, pp. 2727-2731. [172] D. Cox and R. Leck, "Correlation bandwidth and delay spread multipath propagation statistics for 910 MHz urban mobile radio channels," IEEE Transactions on Wireless Communications, vol. 23, pp. 1271-1280, November 1975.   

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.24.1-0074265/manifest

Comment

Related Items