Resource Allocation Techniques for High Altitude Platforms

Transcription

1 Resource Allocation Techniques for High Altitude Platforms This thesis is submitted for the degree of Doctor of Philosophy (PhD) at the University of York Konstantinos Katzis Communications Research Group Department of Electronics University of York October 5, 2005

2 Abstract This work examines various ways to deliver efficient and practical resource allocation techniques for a broadband multi-beam High Altitude Platform (HAP) communication system using frequencies in the LMDS band (28GHz) and above. The hypothesis being tested is that HAP communication systems can take advantage of their physical architecture and exploit cell overlap to maximise the capacity of the network while ensuring fairness in the Quality of Service (QoS) across the coverage area. Performance is judged in terms of blocking and dropped call probability against traffic load; particular attention is paid to fairness of QoS. Through both analysis and computer modelling, an attempt is made to accurately determine practical and near-optimal schemes for optimising the capacity of a HAP communication system whilst maintaining a fair service. Fixed Channel Allocation strategies are examined and the network performance is compared with and without exploiting cell overlap. It is shown that exploiting cell overlap reduces the overall levels of blocking and improves the capacity of the system. It also reduces the dropping levels when the aerial platform starts to move. As users in the overlap regions experience higher interfering levels, they can achieve lower data rates. However they enjoy higher trunking efficiency levels due to the increased number of channels available in the area in comparison to the users in the non-overlapping regions who suffer from higher blocking levels. For the first time the issue of fairness in terms of blocking levels and data-rate is addressed in a HAP environment. Users in overlap regions are allocated more channels to fulfil the minimum bits per connection threshold required but could be blocked according to a mechanism developed called Random Acceptance Factor (RAF). RAF controls the flow of channels in the overlap regions in order to redirect channels to the non-overlap regions to reduce the blocking levels but also to improve fairness. Inter-cell handoff has been considered as a way coping with aerial platform movements. Based on the six degrees of freedom the wireless system dropping levels increase (due to handoff) when the HAP moves at higher speeds. Nevertheless the performance in terms of dropping level improves significantly if a combination of handoff and cell overlap is employed. 1

3 Contents Abstract... 1 Contents... 2 List of Figures... 5 List of Tables Acknowledgements Declaration Chapter 1. Introduction Telecommunication Infrastructure Terrestrial Cellular Communication Systems Fixed Wireless Access Systems High Altitude Platform Systems Telecommunication System Characteristics Thesis Outline Chapter 2. Resource Allocation Techniques Channel Allocation Power Control Adaptive Modulation and Coding Chapter 3. High Altitude Platform Communication Systems Introduction High Altitude Platform & Cell Boundaries Fixed Ground Station High Altitude Platform possible movements Azimuth and Elevation Angle calculation Received Power / Interference Calculation Techniques Conclusions Chapter 4. High Altitude Platform Communication Model Design Introduction System Design and Analysis Context Diagram Top Level Analysis First Level of Decomposition Second Level of Decomposition Third Level of Decomposition Traffic Simulation Model Design Conclusions Chapter 5. HAP Communication Model System Design Verification

4 5.1. Verification and Testing Methodology HAP System - Motion Testing Mechanism HAP moves along z-axis HAP moves along x-axis HAP moves along y-axis Yaw: HAP rotates along its z-axis Pitch and Roll: HAP rotates along its x-axis and y-axis respectively Verification of CIR levels Verification of the Traffic Model Conclusions Chapter 6. Fixed Channel Allocation Based Techniques Introduction Fixed Channel Allocation (FCA) Scheme Cell Overlap Fixed Channel Allocation based Schemes Numerical Calculation for Uniform Blocking (NCUB) Sensitivity to numbers of channels per area Summary and Conclusions Chapter 7. Providing fair services whilst exploiting cell overlap Introduction Channel Allocation Schemes Area Based Fixed Channel Allocation Scheme (ABFCA) Region Based Fixed Channel Allocation Scheme (RBFCA) Uniform Fixed Channel Allocation Scheme (UFCA) Uniform Fixed Channel Allocation - II (UFCA - II) Summary and Conclusions Chapter 8. Handoff Techniques and HAP Mobility Models Introduction Handoff Techniques High Altitude Platform Mobility Models Mobility Models and Handoff Steerable Antennas and Handoff for HAP motion correction Handoff Simulation Model Discussion of Results Conclusions Chapter 9. Summary and Conclusions Summary and Conclusions

5 9.2. Novel Contributions Future Work Chapter 10. Bibliography Appendix A - Publications A.1. Full list of publications and invited talk presentations A.2. 5 th European Wireless Conference Mobile and Wireless Systems beyond 3G (EW2005) A.3. Wireless Personal Multimedia Communications Conference (WPMC) A.4. International Workshop on High Altitude Platform Systems (WHAPS) Appendix B Glossary Appendix C Cell Overlap Mathematical Analysis

6 List of Figures Figure 1.1 High Altitude Platforms can either be Airships or Airplanes Figure 1.2 High Altitude Platform From fantasy to reality? Figure 1.3 Example of a two-point wireless communication link Figure 1.4 Snell's Law Reflection and Refraction Figure 1.5 Rain and foliage can cause significant attenuation in microwave links Figure 1.6 Reflected signal arrives through different paths at the receive antenna Figure 1.7 Shadowing loss from high and low base stations Figure 1.8 Signal is scattered and attenuated due to the raindrops Figure 1.9 Scintillation effect more important when having low elevation angles Figure 1.10 Feuillets are causing partial reflections leading to constructive / destructive interference Figure 1.11 Hexagonal Cellular Systems Figure 1.12 Circular shape cells formed in a hexagonal lattice Figure 1.13 Cells in real life Figure 1.14 Interference Sources Figure 1.15 Example for Cluster Size K= Figure 1.16 Example for Cluster Size K= Figure 1.17 Frequency Reuse Pattern Figure 2.1 Fixed Channel Allocation Channel Assignment Scheme Figure 2.2 Channel Assignment in a FCA Static Borrowing Scheme Figure 2.3 Channel Borrowing Scheme and Channel locking Figure 2.4 Channel Assignment in a Dynamic Channel Allocation Scheme Figure 2.5 Users in overlap regions can chose between two or more best stations Figure 3.1 System level view of a HAP Communication System Figure 3.2 Example of an approximately circular and approximately elliptical Cell Figure 3.3 Formation of an approximately elliptical Cell Figure Degrees of Freedom Figure 3.5 x or y-axis flow diagram Figure 3.6 z-axis flow diagram Figure 3.7 z-axis Rotation and the impact on the cells on the ground Figure 3.8 Different approach for the effect of roll from [48] Figure 3.9 Calculation of the azimuth angle with respect to the main beam Figure 3.10 Point C1 is different from point B the Sub Platform Point (SPP) when the HAP is experiencing a pitch effect Figure D View of a high altitude platform system

7 Figure 3.12 Calculation of point K perpendicular to the boresight point Figure 3.13 Calculation of equation of plane B Figure above can be considered as the cross section of the 3D model. Therefore vector CK` is a line on plane B Figure 3.14 Equation of vector AD. At which point does this vector intercept plane B? Figure 3.15 Descriptive diagram Calculation of Azimuth and Elevation Angle Figure 3.16 Representation of θsub and φsub angles Figure 3.17 Representation of θ sub and φ sub angles - Side View Figure 3.18 Representation of θ sub and φ sub angles - Top View Figure 3.19 High Altitude Platform Antenna Mask [51] Figure 4.1 Data Flow Diagram notation [55] Figure 4.2 Example of a Data Flow Diagram Figure 4.3 General view of the simulation code Figure 4.4 First Level of decomposition of the simulation code Figure 4.5 Initial Conditions for the HAP position Figure 4.6 Second level of decomposition of the simulation code Figure 4.7 Platform Orientation model with respect to north Figure 4.8 Cell radius and total coverage area radius Figure 4.9 User uniform random positioning Figure 4.10 Weather Simulation model Figure 4.11 Probability Position model for the HAP [56] Figure 4.12 Third Level of decomposition of the simulation code Figure 4.13 MobLoc (Mobile Location) Array Figure 4.14 Markov diagram - State of exchange Figure 4.15 Markov 2 - State Diagram for modelling the probability of one users being on or off Figure 4.16 Monte Carlo Model - Function Map Figure 4.17 Call arrival and departures within frames Figure 5.1 Hexagonal layout coverage area Users connect to the closest virtual base-station Figure 5.2 Hexagonal layout coverage area Users can connect to any base station provided that their received power is greater than the minimum power threshold Figure 5.3 HAP moving along its z-axis Figure 5.4 Effect on the positions of virtual cells when the HAP moves along its z-axis Figure 5.5 HAP moving along its x-axis Figure 5.6 HAP moving along its y-axis Figure 5.7 Yaw effect - HAP rotates with respect to its z-axis Figure 5.8 z-axis Rotation and the impact on the cells on the ground

8 Figure 5.9 Example of a virtual base-station when HAP rotates with respects to its z-axis Figure 5.10 Left: Pitch effect - HAP rotates with respect to its x-axis, Right: Roll effect - HAP rotates with respect to its y-axis Figure 5.11 Pitch (x-axis) and Roll (y-axis) Angle effects on Cells Figure 5.12 Pitch (x-axis) and Roll (y-axis) cause cells to move in a Hyperbolic manner Figure 5.13 Theta and Phi angle and virtual base stations as been defined in [48] Figure 5.14 Example of a HAP when it rotates with respects to its x-axis Figure 5.15 Definition of internal (R i ) and external (R e ) cell radius in the Simulation Code Figure 5.16 Pitch Effect Rotation with respect to the x-axis for both reference and current simulation (top-view) Figure 5.17 Pitch Effect Rotation with respect to the x-axis for both reference and current simulation (side-view) Figure 5.18 CIR Testing Model - Receiver Positioning Figure 5.19 Cluster size of 1, 4 and Figure 5.20 Centre Cell Received Power Contour Figure 5.21 CIR coverage for cluster size equals Figure 5.22 CIR profile across the coverage area for cluster size equals Figure 5.23 CIR coverage for cluster size equals Figure 5.24 CIR coverage for cluster size equals Figure 5.25 CIR profile across the coverage area for cluster size equals Figure 5.26 Centre Cell Received Power Contour 10 degrees pitch Figure 5.27 CIR coverage for cluster size of 7-10 degrees pitch Figure 5.28 CIR profile across the coverage area for cluster size of 7-10 degrees pitch Figure 5.29 One-Cell Representation Top view Figure 5.30 One-Cell Representation 3D view Figure 5.31 Monte Carlo Model - Algorithm Analytical Diagram Figure 5.32 Error bar results proved wrong in the first approach allowing a user to switch on144 Figure 5.33 New approach on how to allow users to switch-on was successfully verified using the error bar technique Figure 6.1 Examples of cluster size arrangements Figure 6.2 Illustration of a typical 7-cell (beam) scenario Figure 6.3 Cells with no overlap Figure 6.4 Cells considering overlap Figure 6.5 Illustration of Regions 3 cell example Figure 6.6 Illustration of Regions - 7 cell examples Figure 6.7 Illustration of Areas - 7 cell examples

9 Figure 6.8 Percentages of Overlapping Area A, B and C with respect to the normalised radius R i -Circular shape footprints are considered for a 3 cell case scenario Figure 6.9 Circular margin of a 121 cells coverage area Figure 6.10 Inner cell consists of 6B s, 6C s and one region A Figure 6.11 Inner cell consists of 6B s, 6C s and one area A Figure 6.12 Outer cell overlaps with 5 cells Figure 6.13 Inner cell consists of 3BSmall, 2BLarge, 4C and one area A Figure 6.14 Outer cell overlaps with 4 cells Figure 6.15 Inner cell consists of 2B Small, 2B Large, 3C and one area A Figure 6.16 Outer cell overlaps with 3 cells Figure 6.17 Case - 1: Regional Based Channel Allocation Figure 6.18 Case -2: Area Based Channel Allocation with certain restrictions Figure 6.19 Case -3: Area Based Channel Allocation allowing channels to move between all regions of the cell of interest Figure 6.20 Individual blocking probability vs Overlap Radius (top), Total Blocking probability vs Overlap Radius (lower) Figure 6.21 Inner cell consists 6B s, 6C s and one area A Figure 6.22 Individual blocking probability vs Overlap Radius (upper), Total Blocking probability vs Overlap Radius (lower) NOAS Figure 6.23 Maximum Minimum Blocking probability between areas A, B and C (Difference between the maximum and the minimum blocking of area A, B and C for a range of channels) Figure 6.24 Verification of the channel allocation technique. Optimum overlap radius is equal to Figure 6.25 Channel Allocation based on overlapping radius Figure 6.26 MAX MIN blocking function versus normalised overlap radius. Normalised to R e Figure 6.27 Average blocking probability in an overlap cell and in a non-overlap Cell. Normalised to R e Figure 6.28 Blocking Probability Discrepancy between areas Figure 6.29 Area Blocking and Cell Blocking after optimisation of 200 channels per cell and OT of 300 per cell Figure 6.30 Area Blocking and Cell Blocking after optimisation of 300 channels per cell and OT of 450 per cell Figure 6.31 Trunking efficiency plot based on erlangb formula. The number of Channels per Cell in this case is equal to the number of Erlangs per Cell

10 Figure 6.32 Quantisation Error Plot defines the margins the optimisation technique can operate Figure 6.33 (+1 / -1) Channel variation and the effect on the desired blocking probability (which is 0.05 in this example) Figure 6.34 (+1 / -1) Channel variation and the effect on the blocking probability as a percentage of the desired blocking (which is 0.05 in this example) Figure 6.35 Area Blocking and Cell Blocking optimised at R=1.25 for 200 channels per cell 199 Figure 6.36 Area Blocking after optimisation base on 200 channels per cell and OT of 300 per cell Figure 6.37 Area channel allocation (%) vs OT Figure 7.1 Nearest Base Station Scheme - No Overlap Considered Figure 7.2 Area Based FCA Scheme (ABFCA) Figure 7.3 Partitioning areas into regions Figure 7.4 Region Based FCA Scheme (RBFCA). Overlap has improved system performance Figure 7.5 UFCA RAF defines probability of acceptance in area B and C Figure 7.6 Uniform FCA Scheme (UFCA). The blocking probability has been reduced Figure 7.7 Uniform FCA Scheme (UFCA) Blocking Probability in Centre Cell for Different Degrees of Overlap Radius Figure 7.8 Definition of Overlap Radii Figure 7.9 CNIR levels of No Overlap Model and Overlap Model Figure 7.10 Channel allocation flowchart guarantees fair data rate in all areas of a cell Figure 7.11 CNIR and Channel usage plots without cell overlap Figure 7.12 CNIR and Channel usage plots with cell overlap Figure 7.13 CNIR and Channel usage plots with RAF Figure 8.1 Handoff occurs when user moves from one cell to another Figure 8.2 Users are queued for a handoff between point (a) to point (b) Figure 8.3 Two Level Handoff Scheme Case 1 [33] Figure 8.4 Two Level Handoff Scheme Case 2 and 3 [33] Figure 8.5 HAP position cylinder for 99% and 99.9% of time as defined by HeliNet [9] Figure 8.6 Example of y-axis drift Figure 8.7 Diagrammatical representation of Equation [8.3] Figure 8.8 Example of z-axis drift Figure 8.9 Antenna mask example Figure 8.10 Drift on the z-axis scenario Figure 8.11 Example of y-axis rotation (pitch) Figure 8.12 Example of z-axis rotation (yaw)

11 Figure 8.13 Example of random walk HAP movement Figure 8.14 Example of reflection HAP movement Figure 8.15 Steerable antennas can be utilised to correct drift movements and therefore maintain constant coverage area Figure 8.16 Steerable antennas can be utilised to correct drift movements with respect to the z- axis Figure 8.17 Steering mechanism for correcting HAP drift on the z-axis Figure 8.18 Steerable antennas can be utilised to correct pitch and roll movements and therefore maintain constant coverage area Figure 8.19 Users are generated within a circle of 9km and served by 19 cells of 3.15km radius each Figure 8.20 Immediate Handoff Initiation Algorithm Figure 8.21 Identify and Remove Handoff Users Figure 8.22 Add Handoff Users Figure 8.23 Handoff and Dropping probability for drift movement Figure 8.24 Moving fotpints cause dropping Figure 8.25 Cell Boundaries with no overlap Figure 8.26 Cell Boundaries with overlap Figure 8.27 Handoff and Dropping probability with respect to the distance from centre of coverage area Figure 8.28 Blocking Probability Figure 8.29 Pitch HAP movement in three stages Figure 8.30 Handoff and Dropping probability for pitch HAP movement Figure 8.31 Handoff and Dropping probability with respect to the distance from centre of coverage area Figure 8.32 Blocking Probability Figure 8.33 Handoff and Dropping probability for a HAP system as a function of ground speed when subject to different types of motion Figure 8.34 Handoff and Dropping probability with respect to the distance from the centre of the coverage area Figure C.10.1 Illustration of area A, B and C Figure C.10.2 Calculation of Area B Figure C.10.3 Representation of area C Figure C.10.4 θ C is directly related to region C and the size of the radius of the overlapping circle Figure C.10.5 Calculation of θ C angle

12 List of Tables Table 2.1 Modulation and coding figures used to determine capacity (1) Code rate = , rate 3/4 convolutional inner code and 188/204 Reed Solomon outer code, (2) All SNR figures assume BER 10-5 RF bandwidth = 25MHz Table 4.1 Probabilities of state transitions Table 5.1 Testing Mechanism Table 5.2 Positions of virtual base-stations when HAP is at its initial position Table 5.3 Positions of virtual base-stations when HAP moves along its z-axis Table 5.4 Positions of virtual base-stations when HAP moves along its x-axis Table 5.5 Positions of virtual base-stations when HAP moves along its x-axis Table 5.6 Positions of virtual base-stations when HAP rotates along its z-axis Table 5.7 Pointing angle for zero pitch angle for V. Base Station 3 obtain from [48] Table 5.8 Pointing angles after 10 pitch angle for V. Base Station 3 and obtain from [48] Table 5.9 Before and after the 10 pitch angle 7 Cell system based on [48] Table 5.10 Positions of virtual base-stations from reference simulation [48] Table 5.11 Positions of virtual base-stations from reference after 10 pitch effect Table 5.12 Positions of virtual base-stations applying the reference simulation parameters Table 5.13 Positions of virtual base-stations when HAP rotates with respect to its x-axis Table 5.14 Verification of Results before the 10-degree pitch effect Table 5.15 Verification of Results after the 10 pitch effect Table 5.16 Simulation Parameters Table 5.17 Simulation Parameters for one cell Table 6.1 Limits of Overlap for the communication model Table 6.2 List of Areas for various Overlapping Cases Table 6.3 Definitions of Regions Table 6.4 List of simulations performed Table 6.5 Simulation Parameters used Table 6.6 Basic Simulation Parameters CACOS Table 6.7 Basic Simulation Parameters NOAS Table 6.8 Basic Simulation Parameters Numerical Calculation of Channel Assignment Table 6.9 Optimum channel allocation for R= Table 6.10 Basic Simulation Parameters for the Optimisation Technique Table 6.11 Optimisation Technique Performance evaluation tests Table 6.12 NCUB optimisation Technique Case of 200 Channels / Cell Table 6.13 Optimisation Technique Case of 300 Channels / Cell Table 6.14 Optimisation Technique Case of the varying radius

13 Table 6.15 Optimum channel allocation for R= Table 7.1 Simulation Parameters for FCA schemes Table 7.2 Channel Allocated per region in centre cell using RBFCA Table 7.3 α, β and OT Optimum parameters used for overlap radius R= Table 7.4 Default Parameter values used to assess performance Table 7.5 Modulation and Coding figures used to determine capacity Table 7.6 RAF parameters Table 7.7 Blocking levels generated from each scheme Table 8.8 Handoff Scenario Simulation Parameters Table C.1 Three Cell Approach Equation for area A, B and C

14 Acknowledgements I would firstly like to thank the Engineering and Physical Sciences Research Council (EPSRC), QinetiQ, University of York and my parents for funding this work. I am grateful to my two supervisors Dr Dave Pearce and Dr David Grace, for the constant encouragement, and support. Many thanks go to the other members of the Communications Research Group for useful discussions, particularly Dr John Thornton. Finally, I would like to thank my parents Andreas and Maria Katzi as well as my cousin and friend Dr Andreas Georgiou for their patience, love and understanding that have shown me through my postgraduate studies. 13

15 Declaration Some of the research in this thesis has resulted in publications in journals and conference proceedings. These papers are included in Appendix A. All contributions presented in this thesis as original are as such to the best knowledge of the author. References and acknowledgements to other researchers in the field have been given as appropriate. 14

16 CHAPTER 1. Introduction Chapter 1. Introduction 1.1 Telecommunication Infrastructure Terrestrial Cellular Communication Systems Fixed Wireless Access Systems High Altitude Platform Systems Telecommunication System Characteristics Thesis Outline 34 This thesis concerns optimum radio resource management schemes for High Altitude Platform (HAP) systems. Chapter 1 aims to introduce the reader to the cellular technology that HAPs are based on. More specifically, in section 1.1 a brief historical timeline of the communications is being presented. Following, in section 1.2 the terrestrial cellular communication systems are presented to stress the differences and similarities with the HAP systems. In section 1.3 the Fixed Wireless Access (FWA) communication system is presented. This is a terrestrial communication system that delivers similar services to what HAP system is designed to deliver in HeliNet and is also directly related to this work. Then the HAP architecture is presented in section 1.4. This allows the reader to picture the advantages and disadvantages of this architecture before moving to the design and implementation of a HAP communication system. In section 1.5, the fundamentals of wireless communication systems are presented to give the reader the relevant background needed to read through the rest of this thesis. Finally, the main focus of this PhD work is outlined and the hypothesis of this work is defined Telecommunication Infrastructure Telecommunication between people has always been an important asset for promoting peace, trade and generally eases our daily life. Telecommunications can be defined as the long-distance communication. This is a complex word consisted of the Greek word tele which means far off and communications. In the earlier times people used various techniques to communicate from a distance such as smoke signals, drums, light beacons, and various other ways. Communications nowadays however are carried out with the aid of electronic equipment such as the Radio, Telephone, 15

17 CHAPTER 1. Introduction Television and many other electronic devices. The equipment used in a telecommunications system are: a transmitter, one or more receivers, and a channel or means of communication such as the air, water, wire, cable, optical fibre, communications satellite, or some combination of these. The information that is transmitted through the channel can be in the form of voice, coded symbols, pictures, data, or a combination of these Terrestrial Cellular Communication Systems In the early days of mobile communications, the systems suffered from poor performance in terms of blocking probability during the peak hours as well as suffering from poor coverage. The high levels of blocking probability were because of the inefficient utilisation of the channels and the high demands of the customers for this type of service [1]. The poor coverage on the other hand was because of the limited number of stations available in the coverage area and the various propagation effects that limited the service availability (see section 1.5). As a result the mobile phone had to have a powerful enough transmitter in order to be able to communicate with the central antenna tower several kilometres away. Therefore, a very powerful, bulky and battery-thirsty transmitter was required that made the equipment not very portable [2]. Examples of the old mobile technology are Mobile Telephone Service (MTS), Improved Mobile Telephone Service (IMTS) MJ and Improved Mobile Telephone Service (IMTS) MK. These three systems are examples of the pre-cellular technology we know today and were introduced during the mid-seventies. During the last 25 years there have been a series of technological advances in mobile communication systems. The outcome from these advances is improved flexibility and convenience for the consumer and it is all thanks to cellular technology. Cellular technology has been introduced to resolve the problems experienced in the early mobile communication systems that were mentioned above. What was revolutionary about cellular systems was the division of the coverage area into small cells and the ability to reuse frequencies across the coverage area. There were more base stations, which led to a shorter distance between the user and the nearest base station, so only low power transceivers were required that consumed less power. The first widely used cellular mobile phones started to appear in early 1980s. These phones were analogue and they were usually installed in cars due to their considerable size. However, soon after, the first handheld devices became available. These devices were firstly introduced as the American Mobile Phone System (AMPS) in the USA and later in the UK as the Total Access Communication System (TACS) that was derived from AMPS [3]. These and later systems became known as first generation of mobile cellular communications. As the 16

18 CHAPTER 1. Introduction technology advanced and the equipment became cheaper, millions of people were able to buy and use cell phones. The major technological developments of the cellular systems that have followed up to the present day have been made in order to satisfy the increased need for higher data rates and a wider range of services. The first major change was the development of the second generation of mobile communications with GSM (Global System for Mobile communications) where the cellular systems were designed and built based on digital technology. This has given the users digital speech and some faster data capabilities from the first generation. Moving on to the third generation, its data handling capabilities are also based on digital technology but unlike the second generation, they are oriented towards packet switching rather than circuit switching (except for voice calls which are still circuit switched). The strong demand for higher data rates along with the strict rules for more efficient use of radio spectrum services shaped the third generation to what is known to be today, which is to provide broadband, packet-based transmission of text, digitised voice, video, and multimedia at data rates up to and possibly higher than 2 Megabits per second (Mbps). Typical examples [4] of third generation standards are UMTS (Universal Mobile Telephone System), based on W-CDMA technology, CDMA2000 (Code Division Multiple Access), TD-SCDMA (Time Division Synchronous Code Division Multiple Access) etc. UMTS (W-CDMA) standards are being used by countries mainly in Europe. CDMA2000 standards are mainly used in America, Japan and Korea. TD-SCDMA is being developed in China and an operational system is planned by Fixed Wireless Access Systems Fixed wireless access (FWA) communication systems are considered to be part of the personal communication services (PCS). Local Multipoint Distribution System (LMDS)-band frequencies have so far been little used for the intended purpose of providing broadband wireless Internet access in the UK [5]. One reason for this is that the existing wired infrastructure such as Digital Subscriber Line (xdsl) technology can provide more cost-effectively the available bit-rates. However, in more remote areas where there is no existing infrastructure, the use of broadband technologies can be a costeffective way for making provision for this new service. Therefore, FWA systems can be regarded as complementary to the traditional wireline services and can be used in the case where existing wired infrastructure telephone access solutions are impractical, expensive, or temporary. 17

19 CHAPTER 1. Introduction FWA is intended for the fixed or non-fixed residential customers and small offices/home offices (SOHO) that can make use of wireless high-speed Internet access (i.e. web browsing / access) as well as voice services (e.g. real time conferencing) [6]. This is of course assuming that the users are located within the area of coverage of an access point or Base Transceiver Station (BTS). Users can connect to the telephone network via wireless radio links, which are allocated to them when required. However, in the case of the existing wired telephone infrastructure, the users are guaranteed a free channel. Thus FWA performs channel allocation upon request, rather than providing all users with resources to access the network. The links can either be Line-of-Sight (LOS) or Non-Line-of-Sight (NLOS) depending on the position of the user within the coverage area and the type of the FWA system they are using. LOS FWA utilise narrow-beamed antennas which give higher nominal boresight gain than the wider-beamed antennas utilised by the NLOS FWA systems. In addition LOS FWA experiences less interference than the NLOS FWA system due to its narrow-beam antennas. However, in a typical urban / suburban deployment scenario [6] only 30% of the subscribers will have a LOS with the BTS, it is apparent that the rest 70% of the subscribers will have to use the NLOS FWA system. Users using a narrow-beam antenna mast with a LOS connection to the BTS are required to align their rooftop antenna with the BTS. This requires the user antenna mast to be very accurately pointed in order to avoid any sever degradation of the received signal. Typical example of a FWA technology is WiMAX (Worldwide Interoperability for Microwave Access) [7]. WiMAX is a wireless Metropolitan Area Network (MAN) that is implemented based on the IEEE standard. WiMAX could possibly provide up to 50 km (31 miles) of linear service area range and allows connectivity between users without a direct line of sight. It connects IEEE based hotspots with each other and generally to other parts of the Internet. It also provides a wireless alternative to cable and DSL (Digital Subscriber Line) for last mile broadband access. It has been shown [8] that the practical maximum data rates for WiMAX were between 500kbit/s and 2 Mbit/s, depending on conditions at a given site. 18

20 CHAPTER 1. Introduction 1.4. High Altitude Platform Systems A number of current technologies are being developed to provide broadband communications. These are varied in their performance and availability. To name a few: Asymmetric Digital Subscriber Line (ADSL) and its variants, Fibre Optic Cable, Broadband Fixed Wireless Access (BFWA), Mobile Telephony (2 nd and 3 rd Generation), Wireless LAN (WLAN) and Satellites. Each of these technologies has its strengths and weaknesses. In the case of wireless communications, the extensive need for more capacity requires more and more radio resources. As a result alternative ways for enhancing the system capacity are being investigated. Since most of the existing wireless devices use only the lower part of the available spectrum, only the higher mm wave frequencies (above 10 GHz) have been left relatively underused. Using higher frequencies however requires a Line of Sight (LOS) channel. This is one reason why High Altitude Platforms (HAPs) are becoming so popular as a possible way of solving the problem. HAPs are aerial platforms designed to provide Broadband Fixed Wireless access services at high data rates, which due to their height (17-22km) can provide line-of-sight channels over a wide area [9], [10], [11], [12],[13]. They are also very quick to deploy, and show promise of being cheaper than an equivalent terrestrial or space-based system of equivalent capacity. These platforms can be either small planes or airships [14]. Figure 1.1 High Altitude Platforms can either be Airships or Airplanes The advantage of having an airship to carry the payload around is that the payload can be significantly greater than the payload carried by the plane. More specifically, an airship is expected to be able to carry a payload around 1000kg whereas an unmanned airplane only 19

21 CHAPTER 1. Introduction around 120kg [15]. In addition, it is easier to generate and store energy through the large solar panels that will be situated on the top of the airship. The airplane on the other hand is of much smaller size and thus the solar panels carried and batteries will be limited. Thus the power provided to the equipment will be much less than in the case of the airship. The physical architecture of the HAP differs from that of the terrestrial cellular communication systems, as it is not just one fixed base station but many base stations which are effectively colocated on a mobile platform. This makes the entire system much more complicated as the payload will be bigger and heavier and more power will be required for its operation. However, the propagation path of a HAP is less obstructed as there is a direct line of sight view between the HAP and the terrestrial user. Figure 1.2 High Altitude Platform From fantasy to reality? There are many issues to be taken into account when designing an efficient HAP-based system. One of these issues is the interference caused by the antenna s sidelobes [10]. Also, the attenuation or scattering of the signal due to the presence of water in the propagation path is a significant issue to be taken into account [10]. It is also very important to consider the fact that the power available for the operation of a HAP depends mostly on solar energy and fuel cells or batteries: therefore all operations performed on the platform must be optimised to consume the least possible energy. Another crucial issue to be addressed is the station keeping [15] of the platform that could leave some regions of the coverage area without service. All these issues have a significant effect on the optimum resource allocation schemes for the HAP and therefore to this work. 20

22 CHAPTER 1. Introduction HAPs are designed to take advantage of their architecture and therefore can make use of the frequencies that the terrestrial or satellite systems can not effectively utilise [14]. These are the frequency bands that require line-of-sight paths and signals are strongly attenuated by rain [16].[17]. Nevertheless, HAPs can use different bands depending on the service and the location. More specifically, ITU has recently allocated spectrum around 48GHz worldwide [18] and 31/28GHz for certain Asian countries [19]. Spectrum has also been allocated in the 3G bands for use with HAPs [20]. There is an ongoing work ([21], [22], [23], [24])for lower frequencies (3G communications) that could possibly allow HAP communication systems to offer services that existing terrestrial communication systems offer. Spectrum sharing studies related to the coexistence of existing terrestrial communications and HAP communications have been carried out in [19] Telecommunication System Characteristics Wireless communication between two locations requires that both locations have transceivers that communicate via a wireless channel. In this section we will present how these systems work and also address some of the issues regarding their operation. Elements of a communication system A transmitter (point A) and a receiver (point B) are required to communicate via a wireless channel in order to send a message from point A to point B. A bi-directional conversation between the two points implies that transceivers must be used. I.e. both points A and B must be using a transmitter and a receiver. Figure 1.3 Example of a two-point wireless communication link There are several items that must be addressed to ensure high Quality of Service (QoS). The first one is the coverage, the second is the blocking and the third one is the dropping. All these three are defined based on what has the service provided promised to its customers. Thus the coverage is being defined by the provider. So a user within the coverage area would expect to be able to access the service at any time (if possible) provided that there are resources available. The blocking is defined in terms of how many times a user could be blocked before using the service. This clearly depends on the available resources (i.e. available channel) and it should not 21

23 CHAPTER 1. Introduction be location depended. Furthermore, the dropping levels are defined based on how often a user can be dropped while being using the system (more on dropping can be found in Chapter 8). This case should ideally be eliminated as it is considered to be worst than any other case [25]. Provided that a user is located within a coverage area, the service provided must be able to guarantee that the blocking probability of the user remains below a certain threshold (e.g. 5%) and while using the system the user will ideally never be dropped. Path Loss Path Loss can be defined as the loss that occurs when Radio Frequency (RF) waves are transmitted from station A through the air and experience any atmospheric influences and interaction with objects, which can alter the signal received in station B. The free space propagation model assumes the ideal propagation condition that there is only one clear line-of-sight path between the transmitter and receiver. H. T. Friis presented the following equation to calculate the received signal power P RX in free space at distance d from the transmitter: P RX = P TX G RX G TX 2 λ 4πd Equation [1.1] where P TX is the transmitted signal power [26]. G TX and G RX are the antenna gains of the transmitter and the receiver respectively and λ is the wavelength. The free space propagation model represents the communication range as an ideal sphere. In reality however, the received power is subject to fluctuations due to multipath propagation, shadowing and rain attenuation effects being listed in the previous sub-section. Wireless Channels Wireless channels provide the medium through which voice or data can be transmitted. These channels are effectively electromagnetic waves that propagate though space in a similar manner that sound waves propagate due to the movements of molecules in the medium. Maxwell proposed a set of equations that describe the propagation of electromagnetic waves. He found out that the speed of the electromagnetic waves is the same to that of light. 22

24 CHAPTER 1. Introduction Electromagnetic waves have similar properties to sound waves. To name a few, electromagnetic waves can be reflected, refracted, diffracted and attenuated depending upon the medium (material) and the size of the obstacles the waves has to go through. Reflection and refraction usually occurs when a wave hits a surface of an object whose dimensions are much larger than its wavelength and the reflected and refracted waves follow Snell s Law. Figure 1.4 Snell's Law Reflection and Refraction Diffraction on the other hand occurs when a wave hits the edge of an object or an object smaller than the wavelength of the incident wave. Figure 1.4 depicts the case where a wave travels from medium A to medium B of a different refractive index. The result is that the incident wave is split to a reflected and refracted wave. It is common knowledge that an electromagnetic wave generated from a point source fades as it propagates further from its source. A typical example is when having an isotropic source such as light generated from a bulb that fades after few metres. For an input power P IN (Watts), the power density P D (Watts/m 2 ) of this wave at a distance d (assuming an isotropic source) is: P D P = Equation [1.2] IN 2 4πd If a receiver at distance d captures an area A RX experiencing a certain power density, the received power will be a fraction of the input power P IN. The loss of power at the receiver is called spatial attenuation. The electromagnetic wave can further attenuate due to the medium it is traversing. For example, microwaves may be attenuated by water or flora. Attenuation due to the propagation environment is a common problem in wireless systems. That is why it is important to make all necessary calculations so that the link budget can still cope with any sort 23

25 CHAPTER 1. Introduction of natural cause of attenuation. Figure 1.5 depicts the case where the link between the BS (skyscraper) and the Customer s Premises Equipment (CPE) is being attenuated by rain and trees. As mentioned above, electromagnetic waves can be reflected and diffracted. As a result, the signal might follow different paths before reaching its final (receive antenna) destination. Figure 1.5 Rain and foliage can cause significant attenuation in microwave links Figure 1.6 depicts the case where the transmitted signal is being directly received from the CPE through path labelled a. Also, the same signal is received by the same CPE through a reflection by a car labelled b as well as a nearby house labelled c. The signal is said to have experienced multi-path propagation and the overall received signal is the vector sum of all individual multipath signals, which can provide constructive or destructive interference. Since the phases of these components are constantly varying as (for example) cars move, the received signal, even when the receiver is not moving, is constantly changing. This is called multipath fading [27]. The rate of fading is a function of the signals wavelength and the speed of reflecting surfaces. 24

26 CHAPTER 1. Introduction Figure 1.6 Reflected signal arrives through different paths at the receive antenna Expanding networks to accommodate more users can be achieved by placing more base stations and reducing the transmit power of all transmissions so that the levels of interference maintained between the stations are minimal. However, it is a common phenomenon that some base stations are positioned in higher positions than others. This results in severe interference since radio signals from the original base station can be of similar order of magnitude to the signal from the new lower base stations for a considerable distance, as they suffer smaller shadowing losses. Figure 1.7 depicts the case where the CPE can receive signal a and signal b from base station a and b respectively. This phenomenon can sometimes be useful as one base station can act as a backup increasing capacity in the region. Figure 1.7 Shadowing loss from high and low base stations 25

27 CHAPTER 1. Introduction In a high altitude platform environment there are unobstructed Line of Sight (LoS) paths between the platform and the user. However, it is important that attenuation due to clouds and rain are taken into account. In addition, there is the problem of scintillation effects. Rain can cause scattering as well as attenuation of the signal. The effect is frequency dependent and as the wavelength becomes smaller the effect is more severe. HAPs are susceptible to rain as their signal has to propagate through the rain. Figure 1.8 Signal is scattered and attenuated due to the raindrops Figure 1.8 illustrates the case where a signal is being scattered into different directions. As a result, the signal received by the CPE antenna will be attenuated. Also, the scattered signal might interfere with other CPE that happened to be in a different cell but being using the same channels. However it is expected that interference due to rain scatter is unlikely to present a problem primarily because the antenna beamwidths in HAP communication system delivering broadband services (such as the one proposed in CAPANINA [28]) at 28/30GHz and 38/40GHz bands are so narrow, meaning that the scattered power arriving in the sidelobes, will also be attenuated by the rain. Multipath is caused due to the reflections from the buildings, terrain and parts of the HAP. It is unlikely to have multipath due to the refraction in the atmosphere. The effect of multipath on the CPE depends on the choice of antenna type that is used. If for example the antenna used on a train operating at 28/30 GHz is employing narrow beams, multipath is not an issue since all but the direct path will arrive in the train antenna sidelobes. On the other hand, if the antenna 26

28 CHAPTER 1. Introduction employed on a train is a digitally beam-formed antenna, the arrival of interfering radiation on the sidelobes may have to be considered [29]. Terrain multipath occurs when a wave front is reflected by a surface which is smooth, i.e. whose roughness dimensions are bigger than the wavelength. Smooth surfaces tend to be reflectors and many building materials such as windows, the smoother walls and metal beams can be smooth enough to cause specular reflections [29]. Scintillation is another problem that needs to be addressed in the propagation environment. It becomes a problem especially in late summer months, during the early evening time when irregularities in the atmosphere are observed and as a result variations in the amplitude and phase of signals are experienced leading to loss of synchronisation. More specifically, layers of different refractivity are formed due to the atmospheric turbulence. Turbulent eddies in the atmosphere mix air masses with different levels of temperature, pressure and humidity [29]. This causes small random variation in refractive index. Scintillation depends primarily on the atmospheric refractivity index that is frequency independent. Figure 1.9 Scintillation effect more important when having low elevation angles However, the phenomenon becomes more apparent at lower elevation angles for example when the elevation angle θ is less than 5 degrees. At very low elevation angles, scintillation can merge into atmospheric multipath. This causes slower, deeper fades (>10dB) and is the result of partial reflections from atmospheric layers called Feuillets [29]. These reflections can effectively cause constructive and destructive interference. 27

29 CHAPTER 1. Introduction Sudden changes in the atmospheric refractivity index cause rapid changes of the angle of arrival [29]. Figure 1.10 depicts the effect of scintillation where the CPE experiences random variations in angle-of-arrival as well as in amplitude and phase. Tropospheric scintillation is normally regarded as being tolerable on elevation angles above 5 deg at C-band (4 8 GHz), Ku-band (12 18 GHz) and Ka-band (27 40GHz). Figure 1.10 Feuillets are causing partial reflections leading to constructive / destructive interference Excessive scintillation may cause outages due to the lost of carrier synchronisation and as a result give rise to Bit Error Rates (BER) exceeding the minimum specified for that service. Cellular System Technology The idea of cellular technology is to divide space into a series of cells each with its own base station. All users - also known as mobile stations that are situated within the coverage of a cell are ideally expected to connect with the same base station. This means that all users are never more than one cell radius away from their base station [30]. This is a generic idea and it is applicable in all cellular communication systems. Due to the nature of this study, it is necessary to go through the principles of cellular technology and give the reader the insight of how it works. 28

30 CHAPTER 1. Introduction Figure 1.11 illustrates the basic terms used in cellular systems. The uplink is the communications channel from the MS to the BS and the downlink the channel from the base station to the mobile stations. The shape of the cells is often assumed to be hexagonal, as this provides a convenient system for analysis. This shape is chosen to simplify the planning and design of a cellular system as hexagons fit together without any overlap or gap in between them as well as to minimise the total number of base stations to cover a given area. However the shape of the cell can be different depending on how exactly the range of a cell is being defined. For example it can be defined as the area on the ground for which that particular BS provides the best signal. In this case, when the cells are placed in a hexagonal lattice and neglecting shadowing and multipath effects, the cells would be hexagonal as shown in Figure Figure 1.11 Hexagonal Cellular Systems On the other hand if a cell is defined as the area where a BS provides an acceptable service, then the cells will be circular as shown in Figure That is if we again neglect the effect of shadowing and multipath. 29

31 CHAPTER 1. Introduction Figure 1.12 Circular shape cells formed in a hexagonal lattice In a real terrestrial environment the cells are not arranged in a hexagonal lattice. Also, the range of the cells keeps changing due to prevailing conditions such as buildings, hills, trees, cars and all sorts of obstacles, which distort the shape of cells, and could possibly leave regions without service. It is common practice when deploying a terrestrial cellular system to allow the cells to overlap each other in order to ensure adequate coverage and to prevent having any areas without service. Figure 1.13 illustrates what the boundaries of cells in real life would look like. Notice that overlap is usually only allowed between cells not sharing the same frequencies. Also, notice that areas with no coverage could exist. Users in these areas will not be able to use the service. Figure 1.13 Cells in real life The main issue limiting the capacity of cellular networks is the interference caused by the users using the same radio spectrum. There is only a finite number of carrier frequencies, so two users trying to use the same carrier frequency in the same area will interfere with each other. In the 30

32 CHAPTER 1. Introduction following example, the two mobile stations are trying to use the same carrier frequency while being located in adjacent cells. As a result they interfere with each other. Figure 1.14 Interference Sources This co-channel interference is a prohibiting factor for efficient use and reuse of the resources. To increase the spectrum efficiency, the number of available carrier frequencies are often divided amongst the cells so that the cells which use the same frequency form a hexagonal pattern within the larger hexagonal lattice, such that all these co-channel cells have six neighbours. The number of cells with different carrier frequencies in such a pattern is known as the cluster size (K). The following figure depicts the schemes where the cluster size K is 3 and 7. 31

33 CHAPTER 1. Introduction Figure 1.15 Example for Cluster Size K=3 Figure 1.16 Example for Cluster Size K=7 In both examples depicted in Figure 1.15 and Figure 1.16, the set of cells re-using the same carrier frequencies (co-channel cells) are shaded. It is important that the cluster size K must produce a regular clustering scheme such as the examples above. The reason is that to ensure the channels are reused within cells separated at a distance without being too close, disturbing existing users or too far and therefore not utilising the channel spectrum efficiency [31]. 32

34 CHAPTER 1. Introduction The minimum separation in space is also called the channel reuse distance D (see Figure 1.17) for two co-channels (same frequencies) at which there is acceptable interference (CIR), and this can be calculated as follows [32]: D = 3 K R Equation [1.3] where R is the cell radius and K is the reuse pattern (cluster size). So if we consider the total number of channels available in the system to be M, the number of channels in each cell, S, is easily found by M S = Equation [1.4] K with 1 D = 3 R 2 K Equation [1.5] K can assume only integer values 1,3,4,7,9,12 as generally presented by the series ( i j) ij + 2, where i and j are integers [31] [33]. It is apparent that the shorter the channel reuse distance D is, the greater the channel reuse over the whole service area. Figure 1.17 Frequency Reuse Pattern 33

35 CHAPTER 1. Introduction Cellular mobile communication techniques have been widely used in terrestrial communication as well as in satellite communications. Although in both cases, the principle of cellular communications is the same, the actual implementation of the cellular technology is different due to the physical characteristics of the two systems. That is why it is imperative to reconsider the design and implementation of cellular mobile communications from High Altitude Platforms (HAPs) explicitly customised to their physical characteristics Thesis Outline This research is primarily focused on exploiting cell overlap in channel allocation, handoff as well as a way of improving fairness and Quality of Service (QoS). Chapter 2 presents a general discussion and comments on various resource allocation techniques. Chapter 3 presents a detailed description of the HAP architecture that allowed us to develop a deep and thorough understanding of the physical characteristics of a HAP system. Chapter 4 has been dedicated on the design and development of the software tools and techniques used for the implementation of a HAP simulator. In Chapter 5 the verification of this software tool has been presented along with some of the results. In Chapter 6 the concept of cell overlap has been presented along with its analysis. Furthermore, a basic fixed channel allocation (FCA) scheme has been presented based on the theoretical ErlangB traffic model. Its performance has then been evaluated when cell overlap was employed. Based on this simulation model, a number of channel allocation schemes have been examined in order to improve system performance in terms of capacity and fairness and some interesting conclusions were drawn. The lessons learnt from the channel allocation schemes implemented in Chapter 6, are used as a basis for the novel channel allocation schemes presented in Chapter 7. These novel channel allocation schemes were developed on a Monte Carlo based simulation platform. Through this work, a novel channel allocation control mechanism called random acceptance factor (RAF) has been implemented to ensure fair channel distribution between the users irrespectively of their location. Chapter 8 shows that platform movements can be disruptive for the communication services from HAPs. A solution to this problem was to employ handoff to maintain uninterrupted user connections. The system performance in terms of both blocking probability and dropping probability has been compared when employing a simple FCA scheme, an FCA with cell overlap and an FCA scheme that had a number of Guard Channels reserved for the handoff. 34

36 CHAPTER 1. Introduction In Chapter 9 the main conclusions of the thesis are drawn and the novel contributions are identified. Furthermore, ideas for future work are discussed. The Hypothesis Cell Overlap can significantly improve the capacity and quality of service in High Altitude Platform cellular based communication systems. 35

37 CHAPTER 2. Resource Allocation Techniques Chapter 2. Resource Allocation Techniques 2.1 Channel Allocation Power Control Adaptive Modulation and Coding 43 Resource Allocation Techniques (RATs) consist of channel allocation, power control, and adaptive modulation and coding techniques. This work focuses on the channel allocation aspects of the RATs. It is important however, that we briefly introduce all techniques related to RATs so that the reader can obtain a better understanding of RATs. In this chapter a number of channel allocation techniques are presented. The reader can read through this chapter and learn more about these techniques that formed the basis for the development of new channel allocation techniques for High Altitude Platforms (HAPs). More information regarding the existing techniques can be found in the literature that is listed in the reference section Channel Allocation The aim of channel allocation is to manage the channel distribution between the base stations such that the interference is kept minimal and at the same time the traffic demands are fulfilled [25]. Managing the channels can be done on a centralised or distributed basis. In the case of a centralised system, the Base Stations (BS) are required to communicate with a central controller. This gives the advantage of being able to monitor the whole system while deciding which channel and were it should be used. However, the more centralised the system is, the greater is the amount of signalling required. As a result a high packet or call-setup delay occurs and there is a possibility that the system will become unstable. Distributed schemes on the other hand allow each base station to decide and allocate channels based on local knowledge. The more distributed the system is, the less global knowledge is required at each base station and usually the allocation is made to benefit the base station itself. A given radio spectrum (or bandwidth) can be ideally divided into a set of channels either of time or frequency nature. For this work, we have assumed a set of orthogonal channels. These channels are separated with a predefined guard-band in order to avoid interference with each other. All these channels can be used simultaneously while maintaining an acceptable received 36

38 CHAPTER 2. Resource Allocation Techniques radio signal. Since the spectrum is a very expensive and limited resource, it is required to minimise the guard-bands bandwidth and increase the number of the channels. The size of the guard-band is subject to the interference levels generated from a channel to its adjacent channels. The balance between maximising capacity and minimising interference is the primary aim. Channel allocation schemes can be divided mainly into two categories based on the manner in which co-channels are separated. The first one and the most basic is Fixed Channel Allocation (FCA) and the second one is the Dynamic Channel Allocation (DCA). Fixed Channel Allocation Scheme For the FCA scheme, a fixed number of channels are assigned in every cell and these channels can only be used in that cell. The number of the available channels must be the same in all cells. FCA can be considered as a distributed scheme since channel allocation in every cell is performed based on local data. This scheme performs well when the traffic is uniform [25]. However, if for any reason the traffic in one or more of the cells increases, then the blocking probability increases as well. This is the probability that a new user will not be granted a channel. If for example a football match takes place in a stadium that lies within the coverage area of one cell and during the extra time, all fans decide to make a phone-call to their partners saying that they will be late; then this sudden demand of service in a cell might create high blocking probability, whereas neighbouring cells might have idle channels and experience low blocking probabilities. As a result, some fans will not let their partners know that they will be back late celebrating their win in a local pub, leading into huge confusion and misunderstandings. This is clearly unacceptable. 37

39 CHAPTER 2. Resource Allocation Techniques Figure 2.1 Fixed Channel Allocation Channel Assignment Scheme In order to tackle this problem, several follow-ups were developed based on FCA. One type is the Fixed Channel Allocation Static Borrowing (FCASB) scheme [34], which allows highcongested cells to borrow free channels from neighbouring cells. In order to relieve the traffic congestion of these cells, channels are lent to them from nearby cells on the basis that they will cause the least harm in the neighbouring cells. More specifically, a channel will be borrowed considering the frequency reuse distance and the co-channel interference levels. This scheme is useful when the traffic begins to be non-uniform in some cells. Under these circumstances, channels are redirected from cells with lower traffic to the cells with higher traffic in order to maintain an acceptable blocking level. Figure 2.2 Channel Assignment in a FCA Static Borrowing Scheme The disadvantage of this scheme is that when a channel is borrowed, several other cells are prohibited from using it. This is called channel locking [35]. The number of the cells that are 38

40 CHAPTER 2. Resource Allocation Techniques restricted using the borrowed channel, dependents on the reuse pattern K, the type of cell layout and the type of initial allocation of channels to cells. Figure 2.3 Channel Borrowing Scheme and Channel locking To give an example, let us consider Figure 2.3 which illustrates a cellular system of reuse pattern of 7 and which is employing channel borrowing. Supposing that cell 5 in the 1 st ring lets cell 1 to borrow one or more of its channels, then these channels will be stopped from being used (i.e. locked) in all cells with number 5 in the first and second ring that are surrounding the cell that has just borrowed a channel (cell 1). Channel borrowing happens on a temporary basis and lasts only for the duration of a call. Once the call is completed, the borrowed channel is returned back to its original cell and locking restrictions are removed. It is therefore quite likely that during the borrowing time, some users that happen to be walking in a park nearby the football stadium, will not be able to make a phone call. Channel borrowing schemes perform better than the FCA scheme when traffic loads are low or moderate [25]. This is because the number of the borrowed channels is small and therefore the system can cope with any small fluctuations of the offered traffic. However, when traffic loads are high the number of borrowings will increase. As a result the number of channels being locked will also increase preventing channels being used in certain regions. As a complementary technique to the FCA scheme, the DCA scheme has been developed. In DCA all channels are placed in a pool and are assigned to new calls as needed such that the 39

41 CHAPTER 2. Resource Allocation Techniques Carrier to Interference Ratio (CIR) levels are kept minimal. It is a much more complicated technique compared to FCA because a centralised system is required to control and process the signalling overheads. Dynamic Channel Allocation Scheme DCA on the other hand, provides the system adaptability to the traffic and flexibility [36], [37], [38]. In DCA, all channels are available in all cells provided that the interference requirements are fulfilled. Using this scheme it is now possible to service much more football fans than with the channel-borrowing scheme while at the same time the people in the nearby park can also make a call. Figure 2.4 Channel Assignment in a Dynamic Channel Allocation Scheme Although DCA is a very powerful and flexible technique during traffic fluctuations, it becomes weak and unable to perform well in high load conditions, which in this case FCA is most suitable [25]. Another disadvantage of DCA is that it requires a highly centralised system that will be able to manage the channel allocation in all cells. This means that the system will need to monitor all active calls to decide whether to allow any new call into the system. Also the transmitters must be capable of transmitting in all frequencies, which belong to the pool of dynamic channels. This can become an important economic factor when deploying a large scale system with several hundreds of base stations as each of these stations will be costly [39]. It is a common practise that a Mobile Station (MS) is connected to base station with the highest Carrier to Noise plus Interference Ratio (CNIR). This station is usually the one located closer to the mobile user. In real life, in order to achieve full coverage, it is necessary that some of the BS will have common areas of coverage. I.e. they will overlap. Any MS that are positioned into the 40

42 CHAPTER 2. Resource Allocation Techniques overlap area will be able to establish a communication link with two or more BS. So for example the people who are located in the stadium shown in Figure 2.5 will have the advantage to choose between two stations to connect to. It is possible that if the users located in the overlapping areas connect to the least congested BS, then the network capacity can possibly increase. Directed Retry and Directed Handoff Channel Allocation Schemes Several schemes that consider cell overlap have been suggested. One of these is the generalised FCA [26] that allowed a call to be served by any of the base stations that was within range. Also, Directed Retry (DR) [25] and Directed Handoff (DH) [40] are two channel allocation schemes based on cell overlap. More specifically DR directs a new call that failed to connect at a base station, to attempt to connect to another one. DH scheme also takes advantage of the overlapping regions and therefore handovers existing calls from a heavily loaded cell to a light loaded cell. Both DR and DH try to balance the number of available channels over all base stations. Based on DR, Johan Karson et al [26] proposed an enhanced type of DR scheme called Load Sharing (LS) that performs better than the DR. The major difference between the two schemes is the number of subscribers allowed to make use of the overlap when they want to make a call and should they be allocated a channel. In DR users can attempt to connect to another transmitter as far as they can hear it and as far as there are available channels in that cell. In load sharing however, attempt to connect to another transmitter can be made based on a number of states with various restrictions imposed in terms of channel availability in each cell. First state is when the cell has less than l-number of occupied channels and it received originating traffic plus traffic from surrounding cells. The second state is when the centre cell refuses to accept traffic from other cells when the number of occupied channels is higher than l and smaller than n while it does not itself give traffic to the surrounding cells. The third state is when the number of occupied channels is greater than n. In this case existing calls might be switched to neighbouring cells if they are located in an overlap region. However, in the case where the surrounding cells do not allow any traffic from other cells then the system cannot effectively readjust itself by redirecting new calls or existing calls to neighbouring cells. 41

43 CHAPTER 2. Resource Allocation Techniques Figure 2.5 Users in overlap regions can chose between two or more best stations Channel Allocation and Cell Overlap Channel Allocation schemes that considerably improve system capacity are of great importance. However, the schemes that are based on cell overlap are required to ensure that not only the blocking and handoff levels are kept minimal but also the fairness (in terms of blocking and dropping levels as well as communication interruption) is uniform. Xavier Lagrange [26] has investigated the fact that the Quality of Service (QoS) experienced by the users in the network is very different depending whether they are in an overlapping region or not. This fact posed the question whether reducing the QoS variation (between the overlapping and non overlapping areas) through some admission control scheme can possibly result in further improvement in the overall service quality. Lagrange proposed a scheme that restricts the number of available channel for new calls in overlapping areas in order to balance blocking probabilities. It introduces guard channels dedicated for handover and restricted channels to be used in the areas with no overlap. Exploiting the overlapping regions it is possible to accomplish fairness and improve QoS. Significant gain can be achieved through balancing the network quality of service through the service area [41]. From the literature it has been shown that exploiting overlap is proven to be beneficial for a cellular communication system as the system capacity increases. This work has been based on the fact that cells will overlap each other. Detailed analysis of the cell overlap and the channel allocation schemes developed will be presented in the following chapters. 42

44 CHAPTER 2. Resource Allocation Techniques 2.2. Power Control The purpose of channel assignment techniques and power control techniques is to assign radio channels to the potential users, such that the CIR levels are maintained within an acceptable range. The philosophy behind power control schemes is based on the fact that the CIR at wireless terminals is directly proportional to the power level of the desired signal and inversely proportional to the sum of the power of the co-channel interferers. Increasing the transmit power of the desired signal and/or decreasing the power of interfering signals, can allow the CIR level to be maintained within acceptable levels. This technique however is based on opposing requirements since on increasing the transmit power the interference levels caused by the MS will also increase proportionally. The solution for this is to find the optimum trade-off between the signal power levels and the interfering power levels. Another critical issue that imposes the use of power control is the fact that the HAP has a limited number of energy resources. As mentioned before, the sources of energy on the HAP are the sun and possibly a diesel engine and fuel cells or batteries. These two sources must provide adequate energy for the payload as well as station keeping. It is therefore critical that energy must be preserved to ensure normal operation of the communication system. Implementation of a power control scheme will require that all radio transmitting operations must be optimised so the system will consume the least possible energy. It is also necessary to monitor and adjust the transmitted power of the signal, depending on the CIR levels. There are several techniques currently implemented in terrestrial systems [25]. Such techniques can be classified as either centralised power control or distributed power control. Centralised power control schemes require a central controller that has complete knowledge of all radio links. This is ideal for the HAP as the base-stations will be positioned in the same place. In the distributed approach, each wireless terminal adjusts its transmitter s power level based on local measurements. This means that the user s equipment will be monitoring the interference in its vicinity and will be adjusting its power accordingly. This scheme is much simpler and therefore implies a lighter and a less complex payload Adaptive Modulation and Coding Adaptive Modulation and Coding is considered as a very important technique for cellular communication systems. A system that employs adaptive modulation and coding can choose a constellation size for modulation and a suitable coding scheme according to the condition of the channel. The better the channel conditions are, the larger the constellation size that can be supported. For poor channel conditions (due to interference or propagation path attenuation) the 43

45 CHAPTER 2. Resource Allocation Techniques system will choose a smaller constellation size for modulation and a different coding scheme. Various modulation schemes considered in [42] are GMSK, 16QAM, 64QAM etc. This ensures that channel capacity can be maximised for a given CNIR. Table 2.1 lists the modulation and coding schemes considered in [42]. This work has been based on these parameters. Mod. Scheme 64-QAM 64-QAM 16-QAM GMSK Code Rate (1) BW Efficiency (Bits/s/Hz) Bit Rate (Mbps) SNR(dB) (2) Table 2.1 Modulation and coding figures used to determine capacity (1) Code rate = , rate 3/4 convolutional inner code and 188/204 Reed Solomon outer code, (2) All SNR figures assume BER 10-5 RF bandwidth = 25MHz If the channel conditions get better, more points can be used in the constellation or a less powerful code. Optimised power control allows the best modulation and coding schemes to be used and along with an optimised to the system channel allocation scheme, can make the best combination for producing an optimised resource allocation scheme for a HAP. 44

46 CHAPTER 3. High Altitude Platform Communication Systems Chapter 3. High Altitude Platform Communication Systems 3.1 Introduction High Altitude Platform & Cell Boundaries Fixed Ground Station High Altitude Platform possible movements Azimuth and Elevation Angle calculation Received Power / Interference Calculation Techniques Conclusions 74 This Chapter aims to describe in detail the High Altitude Platform (HAP) communication system along with the design and analysis of its simulation model. This involves the aeronautical part of the system as well as the communication part of the system. Communications from a HAP are directly dependent on the aeronautical behaviour of the platform thus is imperative that both must be presented. The aim of this tool was to realistically model a HAP communication system that would allow us to carry on a series of simulations related to the radio resource management aspects of a HAP communication system. Thus, using this simulation tool it has been possible to generate and analyse the results in this thesis. This chapter begins with an introduction on the fundamental characteristics of a HAP system. Following, in section 3.2, the reader can go through the major parameters identified that describe the HAP system along with the concept of cell boundaries. In section 3.3, the major parameters identified for the Ground Station (GS) are presented. Then, in section 3.4 the reader can go through the possible movements that an aerial platform can perform. These are the six degrees of freedom that have been used when investigating the platform movement effect on the received CIR levels presented in Chapter 5, as well as on various handoff techniques presented in Chapter 8. Following, in section 3.5, the calculation of the azimuth and elevation angles of the GS is presented. In section 3.6 a number of techniques is presented to calculate the received power. Finally the conclusions are presented in section Introduction As mentioned before, High Altitude Platforms (HAPs) are airships or planes, which will operate in the stratosphere, at 17-22km altitude. Such platforms will have the ability to be deployed 45

47 CHAPTER 3. High Altitude Platform Communication Systems within a short period of time and have the potential to provide Broadband Fixed Wireless (BFW) access services to a large number of users over a wide area [26]. HAPs achieve high capacity by using a large number of wireless transceivers, each using a directional antenna to create cells on the ground. These transceivers are co-located on the platform and they offer a line of sight communication to a geographic service area of approximately 60km diameter [43]. Before presenting the simulation model for the High Altitude Platform (HAP) communication system, it is necessary to identify the main characteristics of a HAP. What does it consist of? How do these parts interact with each other and what would be the best way to accurately model the system in software? A systematic design has been employed to assist the design and simulation of a HAP communication system scenario. The system can be divided into two parts that can be analysed individually; as shown in Figure 3.1. The first object to be considered is the HAP itself and the cells oriented on the ground, which are directly related to the position and orientation of the HAP. Secondly, the Ground Stations (GS) are specified. For this work we will be considering fixed users positioned on the ground. Figure 3.1 System level view of a HAP Communication System Sections 3.2, 3.3 and 3.4 analyse in detail the three parts of the system mentioned before. 46

48 CHAPTER 3. High Altitude Platform Communication Systems 3.2. High Altitude Platform & Cell Boundaries The HAP is the heart of this communication system. All the base stations are co-located on the HAP making it easy to maintain and control. At this stage it is necessary to consider all eventualities that might cause dysfunction of the communication services. These eventualities are mainly the interference caused by the sidelobes of the antennas, the weather conditions, the restricted amount of energy and finally the various movements of the HAP while floating / flying in the air. As will be shown later, the first consideration for the simulation is the various movements of the HAP. This is because all the parameters that define a communication link between the HAP and the user varies when the HAP moves from one position to another. Examining the HAP as complete system, it is possible to describe its state using a number of parameters. The parameters are divided into two categories: HAP and GS parameters. HAP parameters are related only to the HAP, whereas the GS ones are related to the users located on the ground as illustrated in Figure 3.1. The parameters listed below will be considered when developing the simulation model. The major parameters that have been considered when describing the HAP are: 1. Transmitting and Receiving antenna Characteristics Steering / Non Steering Antenna Mask Antenna Gain Transmitting / Receiving Power 2. Carrier frequencies that will be assigned for uplink and downlink 3. Channel assignment scheme used 4. Modulation schemes used 5. Position of the platform X co-ordinate (Sub Platform Point SPP) Z co-ordinate (Sub Platform Point SPP) Y co-ordinate (Sub Platform Point SPP) Roll Angle θ Roll, θ Pitch and θ Υaw angles refer to the rotation of the HAP Pitch Angle with respect to x, y, z axis respectively. Yaw Angle θ North angle (Angle of the antenna on the HAP with respect to north) 47

49 CHAPTER 3. High Altitude Platform Communication Systems φ angle (Angle of the antenna on the HAP with respect to the vertical line pointing perpendicular to the ground) 6. Cell Boundaries as a function of the position of the platform: Cell boundaries can be defined in two ways. The first one is that a cell size and shape is defined as the coverage area of an antenna set by a minimum received power threshold required at the edge of the cell. The footprint that is created on the ground from the main beam is not hexagonal (as depicted in some of the figures) but approximately circular or approximately elliptical (egg-shaped) depending on the type of the antennas employed (symmetric or asymmetric beam antennas respectively), on the HAP and the orientation of the HAP. This is more obvious when having symmetric beam antennas pointing away from the Sub Platform Point (SPP), i.e. the φ angle is significant. Figure 3.2 Example of an approximately circular and approximately elliptical Cell For example this can be explained with the use of a torch. Imagine a torch pointing toward the ground from height h with an angle θ with respect to the vertical to the ground line. In the case of the HAP communication system it may be important to provide beams with a circular footprint instead of approximately elliptical for two reasons: firstly it is easier to tessellate the frequency reuse plans when modelling the system and secondly, the system becomes fairer by equalising the average power on the ground. However, in order to achieve circular footprints it 48

50 CHAPTER 3. High Altitude Platform Communication Systems is necessary to have an elliptical beam. Having antennas with asymmetric beams can be more complicated to design for example there may be a requirement for an elliptic beam primary feed, or for an asymmetric lens [44]. Depending on the carrier frequency, which influences the preferred antenna design, costs and weight may thus be affected. Yet when using antennas with a circular beam, it is inevitable that some of the beams footprints will have an approximately elliptical shape. This is even more evident when considering cells near the edges of the coverage area. Figure 3.3 depicts the case of an approximately elliptical cell that is far from the SPP. Figure 3.3 Formation of an approximately elliptical Cell The second way for visualising the cell boundaries (i.e. the size and shape of each cell) is as the area on the ground that is geographically closest to a particular cell (point of incident of the main beam), in which case cells are considered as hexagonal except the ones in the outer ring which are defined based on a minimum received power threshold. For this work, the boundaries of all cells are defined based on a minimum power received threshold. This allowed us to exploit overlap and quantify it in terms of units of area in order to develop various channel allocation schemes. 49

51 CHAPTER 3. High Altitude Platform Communication Systems 7. Propagation Path: This part of the communication system is defined by various natural parameters such as: Attenuation due to: Rain Distance Vegetation Oxygen Water vapour and cloud attenuation - Scintillation Shadowing (only applied for ground mobile stations) Noise due to man-made effects, thermal noise of the equipment used and the sun Some of the factors above will be further considered in order to define the level of attenuation of the received signal at the Ground Station Fixed Ground Station A Fixed Ground Station (FGS) is defined as a ground transceiver that is characterised by a number of parameters. These parameters define the communication link established between the FGS and the HAP. Similarly to the HAP parameters, the parameters for a FGS are the following: 1. Transmitting and receiving antenna characteristics: Steering / Non Steering Antenna Mask Antenna Gain Transmitting / Receiving Power 2. Carrier frequencies that will be assigned for uplink and downlink depending on the channel assignment. 3. Position of the subscribe equipment x co-ordinate z co-ordinate y co-ordinate 50

52 CHAPTER 3. High Altitude Platform Communication Systems θ Angle: This is defined as the elevation angle as defined by equation See also Figure φ Angle: This is defined as the azimuth angle as defined by equation See also Figure The main problem with the transmit power is the solid state power amplifier constrains. Solid State Power Amplifiers (SSPAs) in the millimetre bands are limited to a maximum power output of few watts [45]. Nevertheless, it is important that transmit power must be used in such a way to save batteries in the cases where mobile transceivers might be used: e.g. a mobile TV studio sending information back to the main studio or a soldier transmitting data back to the headquarters through a HAP High Altitude Platform possible movements Having presented all the parameters of the system we are in a position to analyse the possible movements that the HAP can make. These movements have a direct impact on the position of the cells on the ground and therefore on the communication system. As stated before, a HAP can either be an airship or a plane, however the number of possible unique movements that can be combined together and create another movement will remain the same as we are referring to a flying object. In the following analysis, the HAP is assumed to be an airship. The number of the unique movements is the 6 degrees of freedom. A detailed description of each of these movements is presented in Figure 3.4. The 6 degrees of freedom are further analysed in the following sections. Interpretation of the movements in terms of a programming language was essential for implementing the code. The software model devised formed the simulation platform used for generating any future results. As will be seen later in the thesis, the simulation tool devised has been particularly flexible when modelling various Channel Allocation schemes, Handoff mechanisms and platform movements. 51

53 CHAPTER 3. High Altitude Platform Communication Systems Figure Degrees of Freedom 1 2 Shifting along z-axes. As a result, the cells Shifting along x-axes. As a result the cells on the ground will be enlarged when height on the ground will be shifted in a ±x increases and shrunk when the height direction. decreases. 3 4.Yaw Shifting along y-axes. As a result the cells on the ground will be shifted in a ±x direction. Rotation along z-axis. Platform Rotates while its height remains constant. This causes the rotation of the cells in a circular pattern. 5. Roll 6. Pitch Rotation with respect to y-axis again but this time the effect on the cells is distortion and shifting in the new direction. SPP remains constant. This is similar to the previous case but the rotation takes place on the x-axis instead. Shift along x-axis The HAP moves along x-axis shifting all the cells towards the HAP. The movement can be modelled as part of the simulation tool by adding or subtracting the shifting from every x- position of the cell boresight. This is performed in real time as the simulation code is designed to accept new positions of the HAP. Movement on the x-axis and y-axis can be considered to be 52

54 CHAPTER 3. High Altitude Platform Communication Systems similar as both are performed on the same plane. It is therefore possible to use the same technique to model the y-axis movements. Figure 3.5 illustrates the algorithm that performs the movement with respect to the x/y-axis drift movement. New XSPP / YSPP Position Compare New SPP Position with Old Position The same Do Nothing Not the same Recalculate Cells Positions Cells Position Update X / Y cells positions Figure 3.5 x or y-axis flow diagram Shift along z-axis As the HAP moves up and down (i.e. along the z-axis), the cells expand and shrink. One way to model this movement is to multiply all the positions (centres of the cells) by a ratio as defined by the new z-position and the previous z-position of the HAP. Figure 3.6 illustrates the algorithm that performs the z-axis drift movement. New ZSPP Position Compare New SPP Position with Old Position The same Do Nothing Not the same Recalculate Cells Positions Cells Position Update X,Y cells positions Figure 3.6 z-axis flow diagram 53

55 CHAPTER 3. High Altitude Platform Communication Systems Rotation with respect to the x, y and z-axis of the HAP Before proceeding to the description of this task, it is necessary to list a number of assumptions that can be made for these three cases. First assumption is that the HAP keeps its Sub Platform Point (SPP) constant when there is a change in one of the roll, pitch or yaw angles. Secondly, these angles are defined with respect to the x-axis (for pitch), y-axis (for roll) and z-axis (for yaw). Thirdly, the boresights of the antennas (i.e. the vectors defined from the position of the HAP and every centre of the cell) are assumed to be the vectors to be rotated with respect to any of the three axes. The rotation is performed using a matrix especially used in 3D graphics in computer games. This matrix rotates any vector about an arbitrary axis through the origin [46] [47]. x y z ' v ' v ' v 2 tx + c = txy sz txz + sy txy + sz ty 2 + c tyz sx txz sy x tyz + sx y 2 tz + c z v v v Equation [3.1] Where, ' x v, ' y v and ' z v are the components of the rotated vector x v, y v and z v are the components of the vector before rotation x, y and z are the components of a unit-vector along the x, y and z axis respectively relative to the ground. s, c and t are: s = sin(θ ) Equation [3.2] c = cos(θ ) Equation [3.3] = 1 cos( θ ) t Equation [3.4] As mentioned above, θ is the angle with respect to the arbitrary axis of rotation. The advantage of this matrix is that it can perform any kind of rotation and at the same time reduce the mathematical complexity and processing power required of the computer. This is because fewer calculations are required than any other method [47]. 54

56 CHAPTER 3. High Altitude Platform Communication Systems Performing the calculation of the new positions of the vectors was not just enough to move on to the next stage of this simulation tool. It was necessary to verify that the results were actually correct. The rotation of the vectors is not just simple displacement to a new position, but as we will see later, the boresights are performing some complex rotations. Results needed to be considered carefully and be compared with the work of the others. For this purpose, some predictions were made before actually simulating the model. For example, when the HAP rotates around z-axis, the centres of the cells would be expected to rotate in a similar manner. z-axis rotation - Predicting the results This is perhaps the simplest rotation among the three of them. The HAP simply rotates having as centre of rotation the z-axis of the HAP, i.e. it rotates around itself. Therefore, it would be expected that the cells would rotate cyclically around the SPP point (z-axis relative to the HAP and not the z-axis relative to the origin). Figure 3.7 z-axis Rotation and the impact on the cells on the ground Additionally, it is possible to calculate the position of each centre of the cell with respect to the north direction. As previously defined angle ϑ north is the angle created between the line representing the direction of the boresight of every beam from the HAP and the line pointing north (all lines projected on the ground and measured in an anticlockwise manner). This angle can be calculated by keeping a centre of a cell as a reference so every time there will be a rotation with respect to the z axis, the cells will have a new angle with respect to the north 55

57 CHAPTER 3. High Altitude Platform Communication Systems direction. For simplicity, the centre of one of the cells (preferably at the edge of the coverage area) is set to be pointing towards the north. x, y-axis rotation - Predicting the results For these two rotations, the resultant displacement of the cells on the ground cannot simply be estimated as for z-axis. Here, the cell s trajectory describes a curve which yields both x and y displacements. This can be again illustrated using a torch beam pointing at the ground, inclined in the plane of the shoulders and swinging the arm fore and aft. This technique [48] as shown in Figure 3.8 performs exactly what the rotation matrix in Equation [3.1] does. However, it is not used due to its complexity but it can be used to compare the results of these two methods (see Chapter 5 Section 5.7). Figure 3.8 Different approach for the effect of roll from [48] 3.5. Azimuth and Elevation Angle calculation In order to proceed with the calculation of the received power, and thereafter with the calculation of the Channel to Noise plus Interference Ratio (CNIR) levels, it is necessary to calculate the azimuth and elevation angle for every user in respect of every cell. 56

58 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.9 Calculation of the azimuth angle with respect to the main beam Figure 3.9 can be used as a visual aid drawn in order to calculate the azimuth and elevation angle of a user with respect to the boresight of the antenna on the HAP. Point A represents the position of the HAP and point D the position of the user. Point C represents the point of incident of the boresight for one of the antennas on the HAP and point B represents the SPP point as illustrated in Figure 3.9. If the HAP is stable (i.e. no roll or pitch) then point B or sub-platform point (SPP) must be the same as the centre of the centre cell C1. When the HAP performs pitch or roll (for example pitch as shown in Figure 3.10) then C1 is no longer the same as the SPP point. 57

59 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.10 Point C1 is different from point B the Sub Platform Point (SPP) when the HAP is experiencing a pitch effect As you can see in Figure 3.9 the angle which requires to be measured, is actually the angle that is formed on the plane perpendicular to the direct beam line (AC). More specifically the azimuth angle can be defined from the angle subtended between lines CD` and CK`, and for the elevation the angle subtended between lines AD and AC. In order to calculate the azimuth and elevation angle, a number of steps need to be followed. 58

60 CHAPTER 3. High Altitude Platform Communication Systems Figure D View of a high altitude platform system Observing Figure 3.11, line AC represents the main beam (boresight) as it is transmitted from the antenna located on the HAP to a point (the centre of a cell) realising a virtual cell. In practice, the shape of the cell depends on the way we realise the cell boundaries. As mentioned in section 3.2 point 6 Cell Boundaries, the way that the cell boundaries are defined is by setting a minimum received power threshold. Therefore the boundaries of the cell will not be hexagonal as presented in Figure 3.11 but either approximately elliptical or circular [48]. (See also Figure 3.1, Figure 3.2 and Figure 3.3) Distance between the Platform and Ground Let the platform position be A (x spp, y spp, z p ) where x spp, y spp define the sub platform point (SPP) of the platform on the ground and z p is the height. This way it is possible to calculate the distance of the platform from the ground (h o ) even when the HAP is not vertically aligned with the ground (i.e. when point B is not the same point as C1 Figure 3.10). In any case the following expression defines the height of the HAP from the ground. h o = z p 59

61 CHAPTER 3. High Altitude Platform Communication Systems Distance between Platform position and n Cell In order to calculate the distance AC (Figure 3.11) between the platform and the centre of a cell which the platform antenna is pointing at, the following equation is derived. Cn 2 [ ] 2 2 ( z ) + ( x x ) + ( y y ) AC = h = Equation [3.5] p spp Cn spp Cn It is also useful to calculate the ground distance (BC) as defined on the xy plane (ground). Note that n denotes the n th cell. At this point, it is useful to remind the reader that these calculations are required for calculating both azimuth and elevation angles which will then be used for determining the received power from a user (at a given position) on the ground with respect to any base station located on the HAP. Section 3.6 presents how these two angles are used to determine the received power on the ground. Distance between point B and centre of n-cell The ground distance (gd) (or BC looking Figure 3.11) between point B and the centre of any cell on the xy plane can be calculated using the following equation. Cn ( x x ) + ( y y ) 2 spp 2 Cn spp Cn BC = gd = Equation [3.6] Angle between h Cn h o The angle θ cn ( BAC look at Figure 3.11) can be calculated as follows, h = arctan h o θ Cn Equation [3.7] Cn Vector Representation of AC Another useful calculation to perform is to express the boresight as a vector. This will also be useful when performing various rotations during the simulation. Therefore, AC can be expressed as, 60

62 CHAPTER 3. High Altitude Platform Communication Systems AC = OC OA Equation [3.8] Now, let BK be a vector intercepting AC at an angle of 90 degrees. Then vector AC will be divided into two parts. The length of the first part ( AK ) will be denoted as µ whereas the second part ( KC ) as λ. Therefore, point K can be expressed as follow: ( λx + µ x ) C A x K =, λ + µ ( λy + µ y ) C A y K =, λ + µ ( µ z ) A K = z Equation [3.9] λ + µ Figure 3.12 Calculation of point K perpendicular to the boresight point BAC = 90 ACB ABK = 90 BAK Since BAC = BAK then ABK = ACB and KBC = BAC. 61

63 CHAPTER 3. High Altitude Platform Communication Systems The length of AC was previously calculated (also denoted gd ) as well as angle BAC (θ cn ). Cn Using these two values it is possible to calculate angle and λ and µ. ACB and therefore the length of BK ACB = 90 BAC Equation [3.10] Then, BK = cos ( KBC) BC Also, µ = cos( BAK) AB Equation [3.11] and, λ = cos( KCB) BC Equation [3.12] λ and µ can be calculated simply using Equation [3.11] and Equation [3.12]. However, both equations can been rewritten in a simpler format without the cosine function in order to reduce the processing power required to perform them in a programming language such as MATLAB or C++. It is important in a large and complex simulation model to optimise the code as much as possible in order to reduce the simulation time required. In this case, both equations would have to be performed several times (for all users and base stations) taking a considerable amount of time. It has therefore been useful to rewrite Equation [3.11] and Equation [3.12] to Equation [3.13] and Equation [3.14] respectively that did not contain the cosine function. AB AC From Figure 3.12, cos ( BAC ) = and cos ( ) two angles are equal to can be written as: ACB = BC AC. As mentioned before, these BAK and KCB respectively. Therefore, the following expressions AB AC µ = µ = AB 2 AB AC Equation [3.13] 62

64 CHAPTER 3. High Altitude Platform Communication Systems BC AC λ = λ = BC 2 BC AC Equation [3.14] Now it is necessary to find the equation of plane B which is perpendicular to the boresight vector AC. According to the equation of a plane, it is necessary to know a point in the plane and a vector normal to the plane [49]. In order to do this, it is true by construction that distance KC = BK`. Figure 3.13 Calculation of equation of plane B Figure above can be considered as the cross section of the 3D model. Therefore vector CK` is a line on plane B. Furthermore, angles BCK ' = KBC = CAB and CBK ' = KBA = ACB Another observation made is that the two sets of angles mentioned above will be equal to 45 degrees when rectangle CKBK` becomes a square. 63

65 CHAPTER 3. High Altitude Platform Communication Systems To derive the equation of plane B (see Figure 3.13), that is perpendicular to the plane of boresight (plane A); it is true by construction that vector the figure above, BK ' and vector KC are equal. From BC = BK' + K' C K' C = BC BK' Since BK ' = KC then, K' C = BC KC Equation [3.15] From this equation it is possible to extract point K` that is important for the estimation of the azimuth angle. Also the unit vector û which is perpendicular to plane B (Figure 3.13) is equal to: BK' u ˆ = Equation [3.16] BK' Finally, the distance of plane B from point B (in this case is considered as the origin) is equal to BK '. Since all the parameters of the equation of plane B are known, it is possible to write down the equation of this plane. BC uˆ = BK' Equation [3.17] where, BC represents a random vector from the origin B to the plane B, û is the unit vector perpendicular to the plane and BK ' is actually the shortest distance from the origin B to plane B. Furthermore, vector on plane B. K ' C can be calculated. This vector is now the reference of the boresight Up to now, the relation between the platform and the cells has been presented. As a next step, it is required to perform similar calculations related to the position of the receiver. This is much 64

66 CHAPTER 3. High Altitude Platform Communication Systems simpler as the only calculation required is the one to find the vector AD (where D is the position of the user). All these are depicted in the following figure, which represents the interaction between platform, cells and receivers. Figure 3.14 Equation of vector AD. At which point does this vector intercept plane B? The question now is At which point does vector AD intercept the ground? The answer is given by substituting vector BC of the equation of plane B with the general vector equation of vector BD : this will give the actual position that the position vector of the user with respect to the SPP point, intercepts plane B. The general vector equation BD can be expressed as follow: BD (x B, y ) + λ B ( x, y ) = Equation [3.18] BD BD Recalling the equation of plane B, BC uˆ = BK' and assuming that point D is the point that vector BD intercepts plane B, then the plane equation can be expressed as follows: 65

67 CHAPTER 3. High Altitude Platform Communication Systems BD ' uˆ = BK' Equation [3.19] From this equation, it is possible to calculate the value of λ. ( x + x λ) x + ( y + y λ) y + ( z + z λ) B BD BK' λ = x [( x x ) + ( y y ) + ( z z )] BD B û x û û + y B BD B y BD û û + z BD û B z û û B BD z û = BK' Equation [3.20] 66

68 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.15 Descriptive diagram Calculation of Azimuth and Elevation Angle 67

69 CHAPTER 3. High Altitude Platform Communication Systems Knowing the value of λ, it is possible to find the position of point D and in extend the vector equation of point D C. The dot product between vectors D' C and K ' C (this vector is the reference of the boresight on plane B) will determine the azimuth angle. Both vectors (and the azimuth angle) are placed on the same plane (plane B). AZIM n D' C K' C = arccos D' C K' C ϕ Equation [3.21] The elevation angle can also be calculated using the dot product between vectors AD and AC. ELEV n AD AC = arccos AD AC θ Equation [3.22] Position D of the receiver is randomly selected and it can be positioned anywhere on the map (i.e. the ground plane) like all other positions (A, B and C). For the simulation it is considered that all positions are either uniformly distributed random positions or just uniformly distributed positions. In section 3.5 the calculations for the azimuth and elevation angle for every user with respect to every cell are presented. The following section shows how these two angles were used for calculating the received power of each user with respect to every base station on the HAP Received Power / Interference Calculation Techniques Both azimuth and elevation angles presented above will vary when the position of the HAP varies. These angles can be interpreted into attenuation levels when using one of the following models: 1. The ITU type antenna Mask [18] 2. Flat sidelobe model [50] 3. Measured patterns [44] 4. Theoretical aperture radiation pattern [51] The first method proposed by ITU is specific to the 47/48GHz band which is effectively a sidelobe power envelope which a given antenna must not exceed. This technique however over- 68

70 CHAPTER 3. High Altitude Platform Communication Systems estimates interference when used as a model for predicting co-channel interference in a cellular system [52]. This is because antenna patterns in practice exhibit regions of nulls. That is why a flat sidelobe model (the second model in the list) can be considered more suitable option when the level is chosen to correspond to the average sidelobe level. Following, the measured pattern model is practically very difficult to be implemented as it will require measurement data for each cell in all 3 dimensions. Finally the fourth model mentioned can be considered as a good approximation for reproducing the peaks and nulls associated with the sidelobe region. However, for this work the flat sidelobe model is considered to be the most suitable as it is simple to implement and it requires less processing power to run. Also, the averaging of the sidelobes is a good approximation which let us exploit cell overlap and in extend model various channel allocation schemes. In [50] it has been shown that elliptic beams (assuming flat sidelobes) offer great advantages in terms of optimised power at the cell edges over the circular beams. With the exception of the central cell, all other cells (in the case of circular beams) are distorted since each cell subtends a lesser angle in elevation. The distortion observed in [50] showed that CIR patterns tend to be pushed radially away from their intended location. Elliptical beams on the other hand provide better geographically defined regions of CIR, which is also higher than the case of circular beams. A large proportion of the PhD work has been devoted on the exploitation of cell overlap and how it could be employed to improve the quality of service (QoS). Employing elliptical beams (and therefore having circular footprints on the ground) with flat sidelobes has been of great use when exploiting cell overlap. This is because of its advantages mentioned above but also because the simple geometry that circular footprints offer. Elliptical Beam Calculation The received power on the ground can be calculated using a model for the antenna radiation pattern, deriving the angular displacements from antenna boresight as a function of ground coordinates. Such an approach is presented in [50] where a main lobe radiation pattern of the form ( θ ) n cos is assumed. Furthermore, directivity can be estimated using the GS antenna pointing angles for φ AZIM and θ ELEV angles as shown in Equation [3.21] and Equation [3.22] respectively to finally derive the received power at each user position on the ground. Based on [50] the following calculations have been used to describe either a circular or an elliptical beam: 69

71 CHAPTER 3. High Altitude Platform Communication Systems n { ( ( ))} { ( ( ))} θ n cos θ cos φ cos θ sin φ φ D = D Equation [3.23] max ELEV AZIM ELEV AZIM where D max is D max = 2 arccos ln ( 2) + 2 arccos n θ n φ Equation [3.24] φ AZIM and θ ELEV are the azimuth and elevation angles of the user with respect to the main beam D = D cos for an elliptic and n θ and n φ are the indices for the curve fits of the form ( ) φ beam. n θ and n φ are used for determining the directivity at the cell edges based on the subtended angles θ sub and φ sub respectively. max θ n Figure 3.16 depicts angle θ sub and φ sub. θ sub is the angle formed between line AD and AD` whereas φ sub is the angle formed between line AE and AE`. In practice, the cell will not be hexagonal but approximately elliptical. For this work however, the footprints are considered circular (when the HAP is in a still position with no pitch or yaw). This is feasible by increasing the beam asymmetry towards the edge of the coverage area and narrowing both azimuth and elevation beamwidths to offset the greater link length [50]. Thus each cell will have different values for θ sub and φ sub. θ sub and φ sub will be equal only when considering the cell formed from the SPP beam, i.e. the cell exactly underneath the HAP (this applies when the yaw and pitch of the HAP are zero). In more detail, the level of ellipsis is defined from these two angles. If for example θ sub and φ sub are equal then the footprint is circular. If θ sub and φ sub are not equal then the footprint is elliptical. This applies for the case where the boresight of the cell is perpendicular on the ground (i.e. the SPP footprint). For the rest of the cells, where the boresight is not perpendicular on the ground, the ellipsis will be obvious even when θ sub and φ sub are equal. In order to have circular footprints for these cells, θ sub and φ sub must be calculated for each case. 70

72 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.16 Representation of θsub and φsub angles For θ sub (side view), 71

73 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.17 Representation of θ sub and φ sub angles - Side View 1 SPP C = arctan h spp R θ Equation [3.25] 2 SPP C + R = arctan h spp θ Equation [3.26] θ = θ 2 θ 1 sub Equation [3.27] 72

74 CHAPTER 3. High Altitude Platform Communication Systems Figure 3.18 Representation of θ sub and φ sub angles - Top View After calculating θ sub and φ sub for each cell, it is possible to find the n θ and n φ values using the D = D cos. Since the power at the edges of each cell is required to be half θ n expression ( ) φ max (-3 db) then n θ and n φ can be calculated using the two equations below: n n θ φ ( 0.5) ( cos( θ )) log log = Equation [3.28] sub ( 0.5) ( cos( φ )) log log = Equation [3.29] sub Carrier to Interference Ratio (CIR) Calculation of the received power with respect to every base station is useful when calculating the CIR levels for a user. In order to do so, it is required to take into account the interference caused by all cells (using the same channel) as the antenna mask illustrated in Figure

75 CHAPTER 3. High Altitude Platform Communication Systems shows that the receiving power levels can be significant even at very large angles from the boresight (i.e. far away from the pointing antenna). ϕ AZIM n Figure 3.19 High Altitude Platform Antenna Mask [51] This will require finding the best link between the user and the HAP and then calculating the interference caused by all other cells. Assuming that everybody is transmitting with the same power, the worst case Carrier to Interference Ratio (CIR) is defined from the following equation: Pa Carrier to Interference Ratio = Pi P i a Equation [3.30] The nominator, P a represents the power received at the customer premises equipment (CPE) from the nominal base station. In the denominator all received (interfering) signals are summed up (P i ) from base stations transmitting on the same channel. It is important to minimise the interfering signals in order to increase the CIR levels. This can be achieved by using antennas with low sidelobes, introducing power control to the system and of course allocate channels based on various criteria such as how many users are affected from this channel allocation Conclusions In this Chapter we have presented the HAP communication system architecture and we have shown how the geometry is being defined. Furthermore, we have presented a number of possible movements (six degrees of freedom) that an aerial platform might perform along and 74

76 CHAPTER 3. High Altitude Platform Communication Systems how these movements were converted into code. Finally the method followed for calculating the received power on the ground is being presented. All these have been used in the simulation tool to generate and validate the results that will follow in this thesis. 75

77 CHAPTER 4. High Altitude Platform Communication Model Design Chapter 4. High Altitude Platform Communication Model Design 4.1 Introduction System Design and Analysis Context Diagram Top Level Analysis First Level of Decomposition Second Level of Decomposition Third Level of Decomposition Traffic Simulation Model Design Conclusions 105 This chapter presents in detail the design and analysis of the HAP communication system model. This model is divided into two parts. The first part is the simulation model designed and implemented to realise a three-dimensional HAP environment in order to examine the various platform movements and to calculate the received power levels on the ground. The second part is the simulation model designed and implemented to generate traffic that the HAP communication system is to serve. It also presents the tools used for the implementation of this simulation model and how these tools have been used to deliver the platform that all simulations presented in this thesis were carried. Using this traffic model in conjunction with the threedimensional simulation platform model, it has been possible to develop various channel allocation and handoff schemes exploiting cell overlap while the HAP performed various movements. The results from the simulations are presented in the following chapters. This chapter begins with an introduction presented in section 4.1. This is to describe the general approach used for implementing the HAP communication model design. In section 4.2, some general software engineering design disciplines are presented. This is to give the reader the relevant background knowledge to understand some of the notation and terminology used when designing parts of the code as shown in sections These sections present in detail how the first part (three-dimensional HAP environment), part of the HAP communication model, is decomposed in terms of processes. This part has been used to realistically simulate the communication aspects as well as the platform movement aspects of a HAP communication model in a three-dimensional environment. Then in section 4.7 the traffic model is presented. This a Monte Carlo simulation model based on an ErlangB type of traffic distribution. Finally, the conclusions are drawn in section

78 CHAPTER 4. High Altitude Platform Communication Model Design 4.1. Introduction Like all simulation software models, the design for this simulation tool must: 1. Produce Useful Results 2. Be flexible for any future changes 3. Be fast and efficient when running 4. Be logical and understandable to the users The initial program was designed with a minimal number of processes in order to reduce the complexity of the system and therefore help to ensure correct operation. The system then gradually became more complex as more parameters and processes were introduced. Despite its complexity the system is still flexible and expandable. Some of these optional changes that can be adopted in the future can be any type of new antennas such as smart antennas and perhaps a three-dimensional motion mechanism for simulating real-time HAP movements. For the implementation of the simulation code, the external MATLAB interface called MEX API was used. MEX stands for MATLAB Executable and MEX-files are dynamically linked subroutines produced from C, C++ or Fortran source code. When MEX-files are compiled, they can be run from within MATLAB in the same way as MATLAB M-files or built-in functions. The external interface functions provide functionality to transfer data between MEX-files and MATLAB, and the ability to call MATLAB functions from C, C++ or FORTRAN code. The MATLAB language works with only a single object type: the MATLAB array. All MATLAB variables, including scalars, vectors, matrices, strings, cell arrays, and structures are stored as MATLAB arrays. These arrays contain the internal data that is being passed to the C/C++ environment to be manipulated using common C/C++ programming techniques. More specifically a declaration of a m MEXArray in the C/C++ code corresponds to the internal data structure that MATLAB uses to represent arrays. The characteristics of the mexarray structure contains amongst other things: The MATLAB variables name Its dimensions (m column x n rows) Its type (integer, float, character, etc) Whether the variable is real or complex. If the variable contains complex numbers as elements, the MATLAB array includes vectors containing the real and imaginary parts. The array being passed into the C/C++ code can be accessed using pointers More than one arrays of different dimension can be passed from MATLAB into the C/C++ environment. After 77

79 CHAPTER 4. High Altitude Platform Communication Model Design processing the data in the C/C++ environment it is possible to pass back the results into an array. As mentioned before, the MEX files written in C/C++ language can be compiled in the MATLAB prompt. This is done by typing cmex and then the name of the file containing the code. The compiler is effectively based on a standard C/C++ compiler that MATLAB incorporates. Any sort of compilation errors are reported back in the MATLAB prompt. The reason for using the C++ programming language and the MEX-MATLAB interface was to achieve better simulation performance with the overhead of programming complexity. From previous work [53] it was shown that using the C++ programming language is more powerful and more flexible than MATLAB itself. MATLAB does not process loops as quickly as C/C++ code does. If using MATLAB programming language on its own for implementing a complex simulation platform such as this one could be processing limited by the performance of the computer used as well as non-flexible in terms of the language itself. If on the other hand the C++ programming language is used on its own, this would have been also impractical and difficult as we would have to plot/view the outcome of the simulation. Combination of C++ and MATLAB code into a MEX file has been of great use since all complex calculations have been performed in C++ and all results were presented in MATLAB with the simulation time being probably as fast as it would be in a C++ stand alone code System Design and Analysis The simulation platform is designed in such a way that it can be equipped with any number (n) - of circular or elliptical beams which point to (n) different locations on the ground. Initially each random movement of the platform was modelled in isolation, i.e. individually, and was considered as a single movement (only one of the 6 degrees of freedom considered at a time) while interference between the cells was ignored. This simplified the simulation and gave us basic results, and confidence that the simulator was behaving correctly. At a later stage the system was modified so that the CIR levels could be calculated. The program can accept the position of the platform and the receiver, and is therefore able to perform a number of calculations to find the received power. The position of the main beam for every virtual cell as well as the calculations for finding the received power levels is recalculated every time a new position of the HAP is entered. 78

80 CHAPTER 4. High Altitude Platform Communication Model Design In order to implement the specification mentioned above, it was necessary to implement the simulation code following some engineering disciplines. This was done using YOURDON WORKBENCH [54], a software engineering tool that allowed us to decompose the complex system into processes and data in order to visualise the interaction between them. This is called a data flow diagram (DFD). The basic notation used to create a DFD is illustrated in Figure 4.1. A rectangle is used to represent an external entity, a circle a process, an arrow (labelled with the name of the data) represents the direction where the data is being transferred and the double line represents a data store. External Entity Represents input or output point of information that resides outside the bounds of the system to be modelled Process Represents the transformed of information that resides within the bounds of the system to be modelled Data Item Represents a data item or a group of data items that move from an entity or a process to another process or a data store in the direction the arrow is pointing. Data Store Represents a repository of data that is to be used to store the data and can be used by more than one processes. Figure 4.1 Data Flow Diagram notation [55] The DFD may be used to represent a system or software at any level of abstraction. DFD can be partitioned into levels that represent increasing information flow and functional detail. A level 0 DFD called context diagram, represents the entire software element as a single self-contained unit. This unit is connected with input and output data indicated by incoming and outgoing arrows. Additional process and information flow paths are represented as the context diagram is further partitioned to reveal more detail. The basic form of a data flow diagram is illustrated in Figure

81 CHAPTER 4. High Altitude Platform Communication Model Design External Entity Input Data Computer Based System Output Data External Entity External Entity Input Data Internal Data Stored Data Figure 4.2 Example of a Data Flow Diagram 4.3. Context Diagram Top Level Analysis The following data flow diagram (DFD) depicts the context diagram of the simulation model. At this stage of the software analysis, the input and output data parameters are examined. As seen in the context diagram (Figure 4.3), the simulation code is to receive a structure of various data types related to a HAP system, and generate other useful information about the power levels, interference levels etc. 80

82 CHAPTER 4. High Altitude Platform Communication Model Design Figure 4.3 General view of the simulation code Now, the simulation code as seen above is divided into a number of processes. The decomposition of the main context node called Power Received Calculation Process into other child nodes is presented in the following section. Note that if necessary, each processes presented may be further decomposed. 81

83 CHAPTER 4. High Altitude Platform Communication Model Design 4.4. First Level of Decomposition Figure 4.4 First Level of decomposition of the simulation code The data flow diagram above depicts two processes. Define Cell Radius written in MATLAB Cell / Fixed Station Process (further decomposed) In order to perform these two processes, it is first necessary to define some of the system parameters. These parameters are based on [56] where the HAP is equipped with 127 antennas at a height of 17-22km and the diameter of the coverage area is approximately 60km (see Figure 4.5). The coverage area is then divided into 127 cells as the number of the antennas on the platform. Each cell will effectively have just over a 3.15km radius assuming that all of them have the same size. The task of the Cell Radius calculation process is to calculate the radius according to the system specification, which in this case is the one mentioned above [56]. Now, the value of the cell 82

84 CHAPTER 4. High Altitude Platform Communication Model Design radius is passed to the Cell / Fixed Station process in order to generate these positions. More on the Cell / Fixed Station process is presented in the following pages. 3.15km A km B 30 km Figure 4.5 Initial Conditions for the HAP position This process is static, which means that it will be invoked only once at the beginning of the simulation. Therefore, it is not critical to consider processing power. The second process for consideration at this level of the software analysis is the Cell / Fixed Station Process. 83

85 CHAPTER 4. High Altitude Platform Communication Model Design 4.5. Second Level of Decomposition Figure 4.6 Second level of decomposition of the simulation code Figure 4.6 depicts four processes. 1. Base_Loc written in MATLAB 2. set_hap_cell written in C 3. set_user_hap written in C 4. CIR Calculation (further decomposed) These processes are the backbone of the simulation model. A detailed description of these processes will now be presented. Base_Loc Process (written in MATLAB) As mentioned before, this code is used to generate the positions of the cells in a hexagonal manner. The positions indicate the centre of virtual cells that are created from the antennas on 84

86 CHAPTER 4. High Altitude Platform Communication Model Design the HAP when transmitting to the ground (cells don't really exist - they are just created from antennas). The input data is: Inputs NoOfBases: This is the number of cells on the ground. It is also the same number of antennas on the HAP. ClusterSize: This represents the cluster size of the system. NoOfSectors: This represents the number of sectors, which is set to 1 for a HAP system. HAP_Cell_Current_Radius: This represents the radius of the cells on the ground assuming that all of them are congruent. It is necessary to set the value at the beginning of the simulation in order to generate the positions of the centres of the cells. Outputs BaseLoc Matrix: This matrix holds a set of parameters for a group of virtual base-stations. The virtual base-stations represent the actual positions where the boresight of the HAP antennas touch the ground and of course where the signal strength will be highest. As will be seen later, these positions are necessary when calculating the received power levels for every user. The matrix consists of four parameters: x, y position of the cell, the theta coordinate of the base and finally the cluster number. The x and y co-ordinates will be updated continuously with respect to the current location of the HAP. set_hap_cell (written in C++) The main purpose of this process was to reformat the original BaseLoc matrix into a more suitable format for the simulation. BaseLoc was initially developed to generate the positions of the base stations for terrestrial systems. Furthermore, the process calculates the total coverage area and converts it into an equivalent circle area. As will be shown later, the dimensions of the circular coverage area are needed in order to be able to randomly distribute the users within it. Inputs BaseLoc array, HAP_Cell_Radius and HAP_Cell_Height. The last two parameters are required for the calculation of the coverage area, the alpha and theta north angles (theta north angle as mention in Chapter 3 HAP Internal Parameters). Defining all the parameters, the process generates a matrix called HAP_Base_Loc. This matrix stores all the information about the position of the cells (x and y coordinates) as well as the cluster number. It also keeps some additional information about the angle with respect to north 85

87 CHAPTER 4. High Altitude Platform Communication Model Design (Theta_North) and about the angle with respect to the horizontal line as seen in the figure below. Outputs HAP_Base_Loc: A matrix, which holds (x, y, z, Cluster, Theta_North, Alpha) Coverage Area: The coverage area will effectively vary as the HAP moves. However, the value of this variable defines the size of the coverage area that the system must provide services to the users. Circle_Radius: For simplicity, instead of having a hexagonal coverage area of 60km (ideally hexagonal), the shape of the area is considered to be circular. The size of the area is such that all users are located within the service area. This is an important parameter for generating the users on the ground (see Figure 5.18 in Chapter 5). The Theta_North angle is mainly used for determining the position of the HAP and the cells (individually) with respect to the north. The following figure depicts how the angle is measured. N W α A E S W Theta N orth N E S C B Figure 4.7 Platform Orientation model with respect to north As will be shown later, the received signal and power level process uses the x and y coordinates in order to calculate the power-received levels for each user with respect to every cell (base station). Also, the positions of the cells are updated every time the HAP moves. It is important to say that this process is static, i.e. is only executed once at the beginning of the simulation to 86

88 CHAPTER 4. High Altitude Platform Communication Model Design generate the initial set of coordinates. Therefore any effort for minimising the processing power required for this process would not be of great importance. set_hap_user process (written in C++) - User Positioning Process This is also a static process and is called once at the beginning of the simulation to randomly define the positions of the users. As before, a number of inputs are required for the process: Inputs Circle_Radius, NoOfUsers Outputs HAP_Mob_Loc (x, Y, Z, CLUSTER, TXPOWER, WEATHER ATTENUATION) The Circle_Radius parameter that was calculated in the previous process (set_hap_cell) is now used to determine the area in which the users can be placed. For simplicity, the total coverage area is assumed to be a circle rather than hexagon as shown in the figure below. Coverage Area Radius Cell Radius Figure 4.8 Cell radius and total coverage area radius The users can then be randomly distributed in this area using a technique [57], which generates uniformly distributed random numbers. The aim is to have users randomly distributed within an 87

89 CHAPTER 4. High Altitude Platform Communication Model Design area as shown in Figure 4.9. This figure represents some initial tests being made using MATLAB. The circles represent the users and the stars represent the virtual base-stations. Figure 4.9 User uniform random positioning The position of the users varies randomly or in a square equidistance grid layout as in Figure 4.9. As mentioned before, the output array of this process will contain a number of parameters such as: X, Y, Z, CLUSTER, TX POWER, WEATHER ATTENUATION As you can see, apart from the position of the user, there is other information contained such as the cluster number, the transmit power and the cloud attenuation. These three parameters are not assigned a value at this stage of the simulation as this routine will be invoked and executed only once (positioning the users). However, during the calculation of the received power, the matrix is updated according to the base the user is connected to (cluster size), the received power levels and the weather conditions. The power received level values vary according to the position of the users with respect to the boresight of each antenna on the HAP. This also defines the CIR levels as each connection will 88

90 CHAPTER 4. High Altitude Platform Communication Model Design interfere with any other connections on the same channel situated in other cells. The weather attenuation factor contributes in the overall attenuation and scattering of the signal. However it will be considered as a separate model due to its complexity [26]. Users will in general have different weather attenuation factors since they have different positions. The following figure represents such a model, where the beams are distorted when passing through clouds or perhaps rain. This is illustrated by changing the colour of the beam. Figure 4.10 Weather Simulation model Received Power Level process - set_hap_cir process (written in C++) The received power level calculation is probably the most critical process, as it will be repeatedly called to perform a number of calculations for every change of the platform position, for varying weather conditions and for variations of the user traffic. Again, because of its complexity, the process can be divided into subroutines as depicted in Figure Inputs HAP_Cell_Loc: As mentioned before, this is an array that contains the positions of the centres of the cells, along with the angles that define the direction of the antennas. The direction of the antennas thus should be calculated with respect to the north. 89

91 CHAPTER 4. High Altitude Platform Communication Model Design HAP_Mob_Loc: This is also an array, which contains information about the position of each and every user. It also contains information about the power levels that the user is transmitting with, as well as the attenuation factor that varies according to the position of each user. HAP_Current_Position: While the HAP hovers in the sky, its position will be changing gradually. As a result all the parameters that define the communication link will also change. In order to present a more realistic model of the HAP communication system, an array of values is used, to define the behaviour of the HAP during its navigation. The HAP_Current_Position array holds the following values: X, Y, Z positions as well as the PITCH, ROLL and YAW angles However, it must be taken into account that the position of the HAP must be specified according to [56] and the International Telecommunications Union (ITU) recommendation [18] for High Altitude Platforms. ITU says that the platform should be stationed within a circle of radius 400m with height variations of ±700m [18]. On the other hand, HeliNet project [56] suggests that the platform should be able to move within the large cylinder (Figure 4.11) 99.9% of the time or the small cylinder at 99% of the time. The large cylinder contains the volume that the HAP must be within in order to provide satisfactory quality of service (QoS), whereas the small cylinder provides better CIR levels since it is nearer to the original sub platform point. The dimensions of the large cylinder are 3km height allowing the z position of the HAP to vary between ±1.5km and 4km radius allowing the x and y position of the platform to be anywhere within a radius of 4km from its initial central position (with respect to the coverage area). The small cylinder has a height of ±0.5km and height of 2.5km. 90

92 CHAPTER 4. High Altitude Platform Communication Model Design 4km 2.5km ±1.5km ±0.5km A 99.9% 99% 22 km C B 3.15 km Figure 4.11 Probability Position model for the HAP [56] In the case the platform is an airplane, it is difficult to apply the ITU specification since the platform location boundaries are small. Nevertheless the HeliNet specification can be used since it is somewhat more relaxed than the ITU specification and is intended to be realistic for a plane-based application [56]. If the platform is an airship then the position can be specified more tightly, perhaps allowing the ITU specification to be used [18]. Outputs MobLoc: This is still the same array described previously. The array is updated while process Power Calculation Process Gain and Categorisation calculates the gain of the receiver, the CIR and finally assigns a cluster number to the user. Losses FB: Contains the losses from each fixed station to each base-station, including weather attenuation. 91

93 CHAPTER 4. High Altitude Platform Communication Model Design 4.6. Third Level of Decomposition Figure 4.12 Third Level of decomposition of the simulation code HAP Positioning Process - set_hap_cir (written in C++) The first process contained within the set_hap_cir() file is the HAP positioning process. This process performs the positioning of the HAP as shown in Figure The process is designed to perform any sort of rotation and/or translation based on the 6-degrees of freedom. As a result the HAP_Cell_Loc array is updated according to the movement of the HAP and therefore the centres of the cells (boresight) have a new set of coordinates. Furthermore, the Azimuth & Elevation angle Calculation process calculates the azimuth and elevation angle for every user with respect to every antenna. These values are then passed to the Power Calculation Process in which the power levels will be calculated. Finally, an array of all the users is created in order to hold the maximum received power level of each user (for every cell) as well as the cluster number and finally interference levels. Once all calculations are completed, the HAP is moved again and all the calculations need to be repeated to get a new set of results. Again it is necessary to mention that the positions of the 92

94 CHAPTER 4. High Altitude Platform Communication Model Design HAP must be generated (and stored) using a separate process, which exploits the navigation behaviour of the HAP. Azimuth and Elevation Angle Calculation process (written in C++) In this process, the azimuth and elevation angles must be calculated. From Figure 3.9 the azimuth and elevation angles were defined according to Equation [3.21] and Equation [3.22] in Chapter 3. In order to perform these two calculations, it was necessary to implement a routine (process), which was called Azimuth and Elevation Angle calculation process. The inputs required for this calculation are the positions of the user, the HAP and finally the centre cell. All the relevant calculations can be found in Chapter 3 page 57 Azimuth and elevation angle calculation. HAP Power RX Calculation Process (written in C++) In order to calculate the directivity of the user with respect to all cells, Equation [3.23] is used. More on Received Power / Interference Calculation Techniques can be found in Chapter 3 page 68. The MobLoc array stores the results from the directivity calculations for every user. Each user is assigned a set of results from the directivity calculations that are used for the CIR calculations and for the allocation of their cluster number. Figure 4.13 MobLoc (Mobile Location) Array Categorisation of the user (finding the best communication link) This is the final task of the main part of this simulation model. This will require finding the best link between the user and the HAP and then calculating the interference caused by all other 93

95 CHAPTER 4. High Altitude Platform Communication Model Design cells. It is required to take into account the interference caused by all cells (using the same channel) as the antenna mask illustrated in Figure 3.19 Chapter 3 shows that the receiving power levels can be significant even at very large angles from the boresight (i.e. far away from the pointing antenna) Traffic Simulation Model Design At this point we present a Monte Carlo simulation model based on an approximation to Poisson traffic. As mentioned in the introduction of this chapter, this consists of the second part of the HAP communication system model. This part of the simulation model is designed and implemented to generate traffic that the HAP communication system is to serve. It also presents the tools used for the implementation of this simulation model and how these tools have been used to deliver the platform that all simulations presented in this thesis were carried. The outcome from this model has been a very useful tool that has allowed us to model different complex channel allocation schemes for HAPs. The main constraint of this model is the processing power that limits how realistic the input parameters can be, for example the number of users, the number of conversations, the number of base stations etc. This code was initially based on a terrestrial model [53] developed to model circuit switched traffic. It has been modified to reflect the characteristics of a HAP communication system. Further details about the development of the code will be presented in this chapter. Using this simulation model, it has been possible to develop various channel allocation schemes exploiting cell overlap. All relevant schemes are presented in the order that they were developed; with an emphasis on the schemes that are novel. More specifically, the mathematical analysis of this model is presented in Chapter 6 whereas the Monte Carlo simulation is used to implement and simulate the Channel Allocation schemes presented in Chapter 7 and the Handoff schemes presented in Chapter 8. Relationship to the ErlangB distribution The output of the simulation should approximate to an ErlangB distribution. Designing and implementing the Monte Carlo model required care. This was to ensure that the Monte Carlo model was performing the same task as the theoretical ErlangB model. The general 94

96 CHAPTER 4. High Altitude Platform Communication Model Design characteristics of an ErlangB model are listed below to illustrate the assumptions made when designing the Monte Carlo model. 1. There is a very large number of potential phone users all of whom use the phone infrequently. So the probability of a new phone call starting is independent of how many users are already talking on the phone. 2. In case where a user is blocked, he /she do not automatically try again to get a line. 3. The probability distributions of the length of a phone call and the time between making phone calls are a negative exponential distributions. 4. The proportion of the time that a user is on the phone is very small. Therefore, the time between making phone calls is much longer than the actual lengths of the phone calls. It is well known that based on these assumptions, we can derive the ErlangB formula [58], which is used to calculate blocking probabilities in telephone exchanges. In the Monte Carlo model we need to consider the state of the telephone exchange in terms of the active output channels. This is depicted into a Markov diagram (Figure 4.14) to represent the state of the exchange. Here, the probability of moving from the state where c channels are occupied to (c+1) channels are occupied is just the probability that one of the users switches on in one short time interval. On the other hand, moving from the state where (c+1) channels are occupied to one where c channels are occupied is the probability that in the time interval, one of the users switches off. In the case where the time intervals are short the chances of more than one user switching off at the same time can be neglected. The following section describes in more detail the Markov diagram and how the equations have been derived based on this diagram. λ and µ represent the arrival and departure rates respectively of calls. More specifically, λ represents the probability that any of the users in the system starts to talk, and µ represents the probability that any of the users in the system stops to talk. 95

97 P(Remain OFF)=1-ε.δt CHAPTER 4. High Altitude Platform Communication Model Design λ λ λ λ λ 0 Channels 1 Channels c-1 Channels c Channels c+1 Channels N Channels µ 2µ cµ (c+1)µ Nµ Figure 4.14 Markov diagram - State of exchange The number of users assumed was sufficiently large to approximate an infinite user pool (as defined in ElrangB). It has also been assumed that there are a number of independent events (calls) of negative exponential distribution of length. In order to get a good approximation of a real situation, each user has been modelled as a Markov process assuming two states; the onstate and the off-state. P(Remain ON)=1-σ.δt P(ON to OFF)=σ.δt P(ON) P(OFF) P(OFF to ON)= ε.δt Figure 4.15 Markov 2 - State Diagram for modelling the probability of one users being on or off The on-state denotes that the user is live, transmitting data or speech. Similarly, for the off-state the user is silent. (σ.δt) denotes the probability of one user moving from the on-state to the offstate and (ε.δt) denotes the probability of a user moving from the off-state to the on-state in small time δt: f is the simulation time step in this simulation. To derive the ErlangB formula, we 96

98 CHAPTER 4. High Altitude Platform Communication Model Design assume δt the incremental time is infinitesimally small. In practical terms it has to be so small that the probability of two things (events) happening at the same time is negligible. The probability to remain in the on state P(ON) is (1- σ.δt) and the probability to remain in the off state P(OFF) is (1- ε.δt). These probabilities are independent of what everyone else is doing and how long this user has been in the state. If δt is not infinitesimally small there could be more than one call occurrence into the system during δt and the ErlangB formula could no longer be applied. The probability σ of a call ending in the next simulation time step (f) (where f is an integer indicating the simulation time unit step and where there are f/δt time intervals in one simulation time step) is: σ ' = f 0 t δt f ( 1 σ δt) σ δt = exp( σ t) f [ exp( σ t) ] = 1 exp( σ) = f 0 0 σ δt Equation [4.1] Therefore, the mean time to occupy a channel is: n= 1 n σ' ( 1 σ' ) n 1 1 = ( simulation time steps) Equation [4.2] σ' For the simulation it has been assumed that the simulation time step f is much smaller than the length of a call and therefore σ σ. This model is implemented assuming that the users will make phone calls rather than sending packets of data. Also, the approximation made here is that the same user cannot switch on and back off within the same time interval. Let the probability a user being in the on-state be P(ON), and the probability of a user being in the off-state be P(OFF). Then consider the queue in equilibrium (i.e. the probability of a user moving from the on state to the off state is equal to the probability of moving from the off state to the on state). The following equations describe the system when it is in equilibrium: ( ON ) + P( OFF) = 1 P Equation [4.3] and the global balance equation: ( ON) σ = P( OFF) ε P Equation [4.4] 97

99 CHAPTER 4. High Altitude Platform Communication Model Design From Equation [4.3] and Equation [4.4], ( ON) + P( ON) σ 1 P ( ON) P = ε 1 = 1+ σ ε Equation [4.5] Therefore, the probability of a user being ON can be expressed in terms of σ and ε as following: ( ON) ε ε = σ = 1 + ε σ + ε σ P Equation [4.6] If we multiply Equation [4.6] by the number of users N, this will give us the offered traffic (OT). Since it is known what σ is (the probability of a user to move from the on-state to the offstate) and what the desired OT is (this value is defined at the beginning of the simulation) then: ε σ + ε OT = N P( ON) = N OT ( σ + ε) = N ε OT σ + OT ε = N ε ( N OT) ε σ = (seconds -1 ) Equation [4.7] OT and OT σ ε = (seconds -1 ) Equation [4.8] ( N OT) From Equation [4.2], 1 σ = and therefore, ε, the probability to move from Mean Call Length the off-state to the on-state will now become: OT ε = Mean Call Length (seconds) ( N OT) To summarise, all the important equations are listed in the table below: Equation [4.9] 98

100 CHAPTER 4. High Altitude Platform Communication Model Design Table 4.1 Probabilities of state transitions Description Probability OFF to ON Equations OT Mean Call Length ε = N OT ( ) Probability ON to OFF σ = Mean 1 Call Length The equations listed in Table 4.1 were used when implementing the code to generate the traffic in the HAP communication system model. The type of the traffic generated was an approximation of an ErlangB distribution. Simulation results are compared with the theoretical ErlangB equation for verification. The verification results are presented in Chapter 5 Section 5.9. Software Implementation of the Markov Model The initial program was based on previous work [53] that was explicitly based on a terrestrial cellular environment. Using this model as a starting point has saved crucial time for my research, the time saved being used to focus more on the important aspects such as the channel allocation, handoff techniques and various other topics. The initial code has been restructured and modified to reduce the complexity of the system, and has been tested to ensure correct operation of the Markov model. The code was further developed in order to incorporate all the characteristics of a HAP communication environment. This was essentially an integration of the three-dimensional HAP simulation model presented in Chapter 5 with the ErlangB based traffic distribution model presented in Chapter 6. The outcome was therefore a three-dimensional Monte Carlo based simulation model that has been extensively used to generate a series of results. Despite its complexity the system is flexible and will be able to easily accept any new functions that might be incorporated in the future. For the implementation of the simulation code, the external MATLAB interface called MEX API was used. The reason for using the C++ programming language and the MEX API - MATLAB interface was to achieve better simulation performance with the overhead of programming complexity. From previous work [53] it was clearly shown that using the C programming language results in much faster simulations than just using MATLAB. 99

101 CHAPTER 4. High Altitude Platform Communication Model Design Queuing model implemented in C++ The queuing model was designed and implemented based on the functions illustrated in Figure Function.1 test() - MATLAB Function.2 evaluation() - MATLAB Function.3 setup Bases() - MATLAB Generate Useful Plots Function.4 Cell Overlap Block () C++ Code Function.5 (Header) Setups C++ Code Function.6 Gateway C++ Code Function.7 Fixed Location C++ Code Figure 4.16 Monte Carlo Model - Function Map The functions above are presented in the same order as when the code is running. As mentioned before, the whole software model is a mixture of MATLAB and C++ programming languages using C++ primarily for the heavy processing time-consuming calculations. Function 1 test Description: This is effectively the main function where the input / output variables exist. The code is written in MATLAB. The input values are all defined in MATLAB whereas the output values have been generated in both MATLAB and C++. The input parameters listed below are the main input parameters used in the simulation. These are the number of users in the system, the total number of conversations that are to be generated (lengthy), the total number of independent runs in MATLAB (runsy_matlab), the overlap radius permitted (R_Overlap), the mean call duration of a call (Call_Duration) and finally the number of conversations (Setup_Length) that required to be ignored in order to ensure that the blocking 100

102 CHAPTER 4. High Altitude Platform Communication Model Design levels are right. Setup_Length effectively ignored the first conversations because their blocking probability would have been effectively zero. This is because the system is still empty. Also the number of users is not set to infinite. The most significant output values were the average blocking (avg_blocking), average offered traffic (avg_ot), the blocking probability matrix (Block_Store) which holds the blocking levels for each cell, the offered traffic matrix (OT_Store) which hold the offered traffic levels for each cell and finally the conversation-log matrix (Data_temp) which holds all relevant information regarding the user the base station information, their position, and the user conversation status (e.g. blocked or non-blocked) etc. Inputs users lengthy runsy_matlab R_Overlap Call_Duration Setup_Length % users in the system % generate a total of one hundred thousand conversations % Number of independent runs in MatLab % Overlap Radius normalised to the external radius % Mean call duration of a phone call % How many conversations to get ignore at the start up Outputs avg_blocking avg_ot Block_Store OT_Store Data_Temp % Average Blocking for all runs in matlab % Average OT for all runs in matlab % Individual value of blocking for every run % Individual value of OT for every run % Matrix that holds the log of all the events happened during a run Function 2 evaluation Description: This is the second function in the hierarchy, which effectively coordinates the rest of the processes and plots the results. It interconnects the MATLAB with the C++ environment. Two other functions are called from within this function. The first one is setup_bases, which generates the positions of the cells. After this, the main C++ function called Cell_Overlap_Block is called in order to perform the simulation itself. All the results generated are then returned back to the evaluation function to be analysed and presented. 101

103 CHAPTER 4. High Altitude Platform Communication Model Design Function 3 setup_bases Description: Designed and implemented to generate a table of cell coordinates based on a hexagonal layout and numbered with the appropriate cluster number depending on the cluster size set. All cells have the same radius and carry a cluster number. Function 4 Cell_Overlap_Block Description: This function is the heart of the model. It generates all the useful results. All the channel allocation, power control and modulation and coding schemes are performed here. It is written in C++ language in order to ensure maximum processing speed. More details about the way this function operates will be presented later in this chapter due to its significance. In particular more attention will be given to the channel allocation algorithms, which are the main focus of this research. At this point however, a general description will be outlined leaving the channel allocation schemes description to a later stage. The way the code works is simple. It is based on simulation time steps (f): small time periods of the same duration that sum up to the total simulation time. The code effectively runs through these frames in sequence, during which phone-calls can be made. During this time, it deals with all the users who want to transmit or stop transmitting. Effectively at the beginning of every simulation time step, it removes all users that want to terminate their phone call, and then it looks whether any of them wants to start a new phone call. Remember that we assume it is impossible for the same user to start-and-stop or stop-and-start a new conversation within the same simulation time step. Because of the way the code has been written, a potential source of error was identified and examined. This occurs when two users happened to start and stop a conversation within the same simulation time step. As a result the simulated blocking was expected to be slightly different from the prediction of the ErlangB model (see Figure 4.17). 102

104 CHAPTER 4. High Altitude Platform Communication Model Design User 30 (1) (2) User 65 (3) (4) User 5 Simulation Time Step Figure 4.17 Call arrival and departures within frames In order to give a clearer picture of the problem, the figure above shows some random users attempting to get in or out of the system within the same frame and using the same channel. In the first case we have two users clash (points 1 and 2), one attempting to stop a conversation and another attempting to start one. In this case user 65 which is trying to get the channel will not be blocked because according to the code, user 30 would have been asked first to release the channel and therefore it will be available for user 65. This is because the code scans through the users in numerical order. For point 3 and 4 however things will be different as user 5 will be blocked, as he will be asked first to attempt to get a channel, and since user 65 is using the same channel within the same frame, then user 5 will be blocked although the channel should have been available for user 5 at that time instance within the same frame. Although this looks to be an obvious source of error the two effects cancel out and as a result this does not cause any problem to the results of the simulation. To explain in more detail, consider the two cases where two users happen to start and finish within the same frame. For any event where one user finishes and another one starts a new conversation within the same frame time the probability is 50:50 whether he starts before the other finishes or he starts after the other finishes. Also the probability of the user (number) being bigger than the other user is 50:50 since the software picks up randomly the users for a conversation. Therefore there are two situations that occur equally often during the simulation. The first one is when a user is allowed to start a conversation although the channel is still occupied and the other one is when the user is blocked although the channel has been freed earlier within the same frame. This has been examined and verified by noting whether these two 103

105 CHAPTER 4. High Altitude Platform Communication Model Design events cancel out each other for different conversation lengths. Increasing the conversation length, we effectively decrease the number of occurrences as the probability of two users starting and stopping within the same frame decreases. This however increases significantly the simulation time. On the other hand, decreasing the conversation time, the number of the collisions increases but the simulation time decreases. Increasing or decreasing the conversation time should not make any difference whatsoever since in both cases the number of collisions cancels out. As mentioned before to provide results in agreement with the ErlangB formula, the simulation time step time has to be small enough that the probability of two events happening within the same simulation step is negligible. Function 5 Header Setups This is the header file for the simulations. It contains the declarations for the global structures, enumerated types and a few general-purpose functions such as table look-up and random number generators. At this point it is worth presenting the random number generator routine, as it has been a crucial part of the Monte Carlo simulation. Random Number Generator Routine The random number generator [57] lives in the header file (function 5) and it is invoked every time a user is asked whether he wants to change state from on-to-off or from off-to-on. This routine shuffles the output to remove low-order serial correlation. More specifically, the code initially generates 16 different numbers and one of them is randomly selected to be the seed. The new generated random number then replaces the number used as a seed and then a new number is picked-up from the 16 numbers to be the new seed. This process is recursive and repeats every time the function is called. Every time it is invoked, it returns a uniform random deviate between 0.0 and 1.0 (exclusive of the endpoint values). The computer clock has been used as the seed to generate random values. As will be seen in the verification section of this chapter, the routine has been examined extensively to ensure uniformity and lack of correlation. Since the probability of a user to move from the off state to the on state is extremely small (of the order of ) a problem would exist if the random generator was sensitive to extremely low values. For example, if a very low number was generated, there was a finite probability, lower than the expected one, that the next value will also be very low. The problem is a tendency that after one small number is produced, the next random number will not also be small. Many random number generators are known to have this weakness. 104

106 CHAPTER 4. High Altitude Platform Communication Model Design Function 6 and 7 These two functions are not of great interest as they are there to assist function 4 to operate. They are part of the C++ code written and they are invoked through function 4. Function 6 contains the gateway and setup routines that parse the inputs to function 4 and reserve all the space required by the structures and arrays for the program and then free them up again at the end. Function 7 places random user stations in a two dimensional environment. It also calculates the losses from all base stations to all of these user stations, and puts the results in a large array which is then used to calculate carrier to interference ratios at a later stage Conclusions This Chapter has shown in detail the design and analysis of the simulation tool used to validate the hypothesis of this PhD thesis. This has been a key fundamental tool used to simulate the behaviour of the platform and its effect on the communications from HAPs. More specifically, this tool has been used to realistically simulate the communication links in a three-dimensional environment. It has also been used to design and implement various channel allocation schemes whilst exploit cell overlap. These schemes were based on an ErlangB traffic model which was part of the simulation platform presented in this Chapter. Furthermore, it has been used when investigating the effect of the platform speed and the type of movement on the handoff and dropping probability in the system. This has been a different approach from [59] that was extensively used in the Communications Research Group University of York. The simulation tool implemented here added more flexibility in the investigation of communications delivered from HAPs. Due to its complexity the simulation tool was developed based on a software engineering case tool called Select Yourdon [54]. It has been important that the simulation tools was build based on these software engineering disciplines in order to make it robust for future updates and upgrades that were required. 105

107 CHAPTER 5. HAP Communication Model System Design Verification Chapter 5. HAP Communication Model System Design Verification 5.1 Verification and Testing Methodology HAP System - Motion Testing Mechanism HAP moves along z-axis HAP moves along x-axis HAP moves along y-axis Yaw: HAP rotates along its z-axis Pitch and Roll: HAP rotates along its x-axis and y-axis respectively Verification of CIR levels Verification of the Traffic Model Conclusions 147 In Chapter 4, the HAP communication system model has been divided into two parts. The first part was the three-dimensional HAP environment model that performs the movements as well as calculated the received powers on the ground. The second part was the traffic model that was used to generate the traffic that the HAP was to serve. In this Chapter, we will be looking at the verification and testing of these two parts. Here the tests that have been performed are presented in order to ensure that all parts of the code behave normally, and provide the expected results. This chapter begins with the verification and testing methodology followed presented in section 5.1. In section 5.2 to section 5.7 the verification of the six types of movements (see Figure 3.4) is presented. Following, in section 5.8, the Carrier to Interference Ratio (CIR) levels are calculated assuming all the receivers are positioned uniformly in the shape of a square grid with respect to coverage area. The results are based on different cluster size numbers such as K=1, 4 and 7. In section 5.9, the verification of the HAP traffic model is presented. Finally, the conclusions are presented in Verification and Testing Methodology Verification of the simulation results is a very important task that needs to be performed at the early stages of the simulation. This is to ensure that the early version, the core HAP model simulator tool was correct before using it to investigate various resource allocation techniques. 106

108 CHAPTER 5. HAP Communication Model System Design Verification For the verification process of the simulation tool, a mixture of mathematical analysis and direct comparison with the simulation has been performed. Also, results have compared where possible with results obtained in the past by the University of York Communications Research Group [56], [10]. The following verification strategy was used to drive the three-dimensional HAP environment simulation model. The structure of the testing mechanism is presented in Table 5.1. Table 5.1 Testing Mechanism HAP IN A STATIONARY STATE No. Name Description 1 Random User Positioning The user position must be checked if it is being randomly allocated. To do so, the average number of users per square unit area multiplied by the total coverage area must be the same with the total number of users initially set in the simulation. 2 Elevation / Azimuth Elevation and Azimuth angles must also be verified performing some of the calculations on paper as well as comparing the results with [48]. 3 Received Power Level Verifying the Azimuth and Elevation angles it is possible to check the received power levels by performing some calculations on paper and applying standard propagation theory [60]. 4 Cluster No Assignment The Users must be allocated a cluster number according to the closest base station. What would be this number? 5 Interference Levels Are the interference levels as we predicted from doing the calculations on paper? HAP MOVING ALONG THE AXIS 6 x-axis 7 y-axis 8 z-axis Observe the positions of the cells based on the new position of the HAP, as well as the new received power levels. The cluster number might change depending on how much the HAP has moved. HAP ROTATING WITH RESPECT TO THE AXIS 9 PITCH (x-axis) The positions of the cells must vary in a hyperbolic 107

109 CHAPTER 5. HAP Communication Model System Design Verification 10 ROLL (y-axis) 11 YAW (z-axis) manner. Results can be compared with earlier theoretical results derived within the communications group [48] for consistency. This is just a simple rotation with respect to the z axis. Therefore the positions of the cell will move in a predictable circular manner. HAP COMBINED MOVEMENTS 12 PITCH angle, YAW angle and ROLL angle, x, y & z axis shift Combination of all these movements will attempt to represent the HAP in full motion. The testing mechanism is based on the HAP_Current_Position mechanism that was discussed earlier (section 4.5, Chapter 4) HAP System - Motion Testing Mechanism To verify the technique used for repositioning the virtual base-stations when the position of the HAP varied, a number of virtual base-stations was defined as the positions of the centres of the cells. This is the point where the boresight of each antenna on the HAP intersects the ground. The verification can be divided into 6 steps based on the six degrees of freedom of the HAP which can be found in page 52. Initially, a set of values was generated when the HAP was stationary at a height of 22km. The following table represents the x, y positions (where the antenna boresights touch the ground) of the virtual base-stations at a height z of the HAP. There are 7 cells (virtual base-stations) formed on the ground in this example. Table 5.2 Positions of virtual base-stations when HAP is at its initial position V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z Plotting these points (look at Figure 5.1) creates a large hexagonal array of 7 cells, which represents the total coverage area. The cells are positioned based on a hexagonal layout in order 108

110 CHAPTER 5. HAP Communication Model System Design Verification to ensure complete coverage in the nominal coverage area while minimising the number of virtual base-stations that can be positioned in the area. Figure 5.1 illustrates a typical hexagonalbased cellular layout assuming that all users connect to the closest virtual base-station km km Figure 5.1 Hexagonal layout coverage area Users connect to the closest virtual base-station For this work however, we have assumed that a cell is defined as being the area on the ground which gets the biggest received power from a base station antenna. Figure 5.2 illustrates a set of 7 approximately circular footprints positioned based on this assumption. This approach has been extended to be used for exploiting cell overlap by allowing users to connect not just to the closest base station but also to any base stations that are within range. Therefore, regions of overlap are formed between cells. More on cell overlap can be found in Chapter 6, Section

111 CHAPTER 5. HAP Communication Model System Design Verification km km Figure 5.2 Hexagonal layout coverage area Users can connect to any base station provided that their received power is greater than the minimum power threshold HAP moves along z-axis Z X Y Figure 5.3 HAP moving along its z-axis Description: As mentioned before, when the HAP moves along the z-axis, the positions and the sizes of the cells of the virtual base-stations will change. It is expected that when the HAP moves upwards, the positions of the centres of the cells and their sizes will expand, whereas when it is downwards they will shrink. This is because the elevation angle (the angle between the boresight and the line from the HAP perpendicular to the ground see Equation [3.22]) is always constant irrespective of the height of the HAP. 110

112 CHAPTER 5. HAP Communication Model System Design Verification Initial Position New Position (Lower) B B SPP Figure 5.4 Effect on the positions of virtual cells when the HAP moves along its z-axis Verification: The new HAP height was set to be 21500m, 500m lower than the current position. As mentioned in page 53 Chapter 3 Shift along z-axis, this process resizes the whole structure of the virtual base-stations on the ground. So, the new position of the virtual basestation is the product of the current position and the ratio between the new and the current height of the HAP. From the simulation, Table 5.3 was generated. Table 5.3 Positions of virtual base-stations when HAP moves along its z-axis V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z Taking as an example the position of virtual base-station 3. Its initial position was (5000, 8660) meters and after the HAP moved 500m downwards it became (4886, 8463) meters. The ratio between the new height and the initial height is, new height height ratio = = = Equation [5.1] current height

113 CHAPTER 5. HAP Communication Model System Design Verification 5.4. HAP moves along x-axis Z X Y Figure 5.5 HAP moving along its x-axis Description: As the HAP slides along its x-axis, the x-positions of the virtual base-stations on the ground will be changing as well. Verification: In order to verify this, the current x-position of the HAP which initially was 0, was then set to be 100 metres in the positive direction of the x-axis. What is expected from this shift is that all the x-positions of the virtual base stations in Table 5.4 will now have 100 metres added to their current values. The following table lists the new positions of the virtual base-stations when the HAP moves 100m towards the positive x-axis direction. Table 5.4 Positions of virtual base-stations when HAP moves along its x-axis V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z The simulation results agree with what was expected and therefore this part of the simulation does not need any further verification. 112

114 CHAPTER 5. HAP Communication Model System Design Verification 5.5. HAP moves along y-axis Z X Y Figure 5.6 HAP moving along its y-axis Description: As the HAP slides along its y-axis, the y-positions of the virtual base-stations on the ground will be changing as well. This is a case similar to the x-axis analysis mentioned before. Verification: In order to verify this, the current y-position of the HAP, which was initially 0, was then set to be 100 metres on the positive direction of the y-axis. What would be expect from this change, is that all the y-positions of the virtual base stations in Table 5.2, will now have 100 metres added to their current values. Table 5.5 Positions of virtual base-stations when HAP moves along its x-axis V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z The simulation results agree with what was expected and therefore this part of the simulation does not need any further verification. 113

115 CHAPTER 5. HAP Communication Model System Design Verification 5.6. Yaw: HAP rotates along its z-axis Z X Y Figure 5.7 Yaw effect - HAP rotates with respect to its z-axis Description: The HAP rotates with respect to its z-axis. This means that the centres of the cells i.e. the virtual base-station positions will move in a circular manner with respect to the SPP point. ϑ shift north North Cell No. n +1 North ϑ north Cell No. n Cell No. n +1 Cell No. n + 2 Cell No. n Cell No. n + 2 Initial position of the HAP Next position of the HAP Figure 5.8 z-axis Rotation and the impact on the cells on the ground Verification: In order to verify this, an example of a virtual base-station position will be considered. The HAP is at a height of 22km and is forming 7 virtual base-stations on the ground. Performing a 10-degree anti-clockwise rotation with respect to the z-axis of the HAP, the following table was generated. 114

116 CHAPTER 5. HAP Communication Model System Design Verification Table 5.6 Positions of virtual base-stations when HAP rotates along its z-axis V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z Comparing the position of virtual base-station 3, before (see Table 5.2) and after (look Table 5.6) the rotation: Before Rotation 10 Degrees Rotation x y Z Y X Y B (3420, 9397) B(5000, 8660) θ 1 θ 2 SPP(0,0) X Figure 5.9 Example of a virtual base-station when HAP rotates with respects to its z-axis The verification of the new position B requires some calculations. Let us consider point B, which has coordinates B(5000, 8660). This means that angle θ 2 is given by: 115

117 CHAPTER 5. HAP Communication Model System Design Verification 8660 θ 2 = arctan = 60 Equation [5.2] 5000 Now, a 10 degrees rotation of the HAP will shift point B to position B. Thus the total rotation angle with respect to the x-axis is 70 degrees. Since the length of SPPB and SPPB is the same, i.e. is equal to 10000m, then the position of B can be calculated as follow: ( R) cos( + ) = cos( 70) m x B = θ 3420 ' = θ Equation [5.3] Similarly, for y-axis: ( R) sin( + ) = sin( 70) m y B = θ 9397 ' = θ Equation [5.4] The results above agree with the results calculated from the simulation model. The same way can be used to calculate the rest of the points of interest. It can therefore be considered that this part of the model operates correctly Pitch and Roll: HAP rotates along its x-axis and y-axis respectively Z Z X Y X Y Figure 5.10 Left: Pitch effect - HAP rotates with respect to its x-axis, Right: Roll effect - HAP rotates with respect to its y-axis Description: The rotation with respect to x and y-axis can be considered identical to each other. The following figure illustrates the expected effects of these two types of rotations, on the positions of the virtual base stations and as a result to the size of the cells. 116

118 CHAPTER 5. HAP Communication Model System Design Verification Z Pitch z Roll X Y y x Hyperbolic shape Figure 5.11 Pitch (x-axis) and Roll (y-axis) Angle effects on Cells It can be clearly seen that in both cases (x and y-axis rotation), the boresight line traces out the arc of a circle, giving a hyperbolic path on the ground. The hyperbolic shape is defined as the curve produced by a plane intersecting the nappe [61] as illustrated in the following diagram. Z X Y Nappe Circular Movement with respect to y-axis Ground Plane Hyperbolic Path Figure 5.12 Pitch (x-axis) and Roll (y-axis) cause cells to move in a Hyperbolic manner 117

119 CHAPTER 5. HAP Communication Model System Design Verification Since both x and y-axis rotation are based on similar principles, the verification of only one (Pitch) will be presented. Verification: In this case, it was considered that the best way to verify the rotation with respect to either x or y-axis, was to compare the rotation results with the results from a different simulation [48] based on a different technique which performs exactly the same task. For the verification of the simulation results, a set of data from [48] was obtained. The comparison could not be performed straight away since the two sets of data were in a different format. In this simulation, HAP_Cell_Loc matrix stores the location for each virtual base-station (which is in fact the position where the boresight of a HAP antenna touches the ground). On the other hand, the data from the other simulation [48] consists of pairs of angles (theta and phi) which define the position of the boresight of each antenna (from the HAP) on the ground. Both tables describe the positions of the virtual base-stations in a different way. It was therefore necessary to convert one of these sets into the format of the other in order to be able to compare them. The conversion was performed in MATLAB. A program was written in order to perform this conversion for all the positions. A simple example is presented below which explains how the conversion was performed. It is worth mentioning that the height of the platform was set to be 17km as this was the value used in [48]. Table 5.7 and Table 5.8 represent a pair of angles before and after the rotation respectively for one of the six cells of the first ring of cells in [48]. These angles are translated into the x and y positions of the cell on the ground with respect to the SPP point. The whole set of results for the simulation reference, is presented in Table Table 5.7 Pointing angle for zero pitch angle for V. Base Station 3 obtain from [48] Theta Phi 60 Table 5.8 Pointing angles after 10 pitch angle for V. Base Station 3 and obtain from [48] Theta Phi

120 CHAPTER 5. HAP Communication Model System Design Verification Table 5.9 Before and after the 10 pitch angle 7 Cell system based on [48] V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 Before the rotation Theta Phi After the rotation Theta Phi Before proceeding to the conversion of the theta and phi angles, it is helpful to illustrate them as being defined in [48]. Figure 5.13 depicts how the cells are numbered and how their centres are defined in terms of theta and phi angle. Z Platform Height: 17km X Y theta Cell 4 Cell 3 Cell 5 SPP Cell 1 phi B(x,y) Cell 2 Cell 6 Cell 7 Figure 5.13 Theta and Phi angle and virtual base stations as been defined in [48] More specifically, theta angle denoted the angle formed between the centre of the cell, the HAP and the Sub Platform Point (SPP). The phi angle is the angle between the centre of the cell, the SPP and the y-axis. Figure 5.13 clearly depicts the two relevant angles used to describe the position of point B (where B is the centre of a virtual base-station). 119

121 CHAPTER 5. HAP Communication Model System Design Verification These two angles must be used to calculate point B (Figure 5.13) in order to convert the results into the appropriate format to compare with this simulation. The same method was used for the rest of the pairs of angles, before and after the rotation. There are two sets of coordinates for the comparison of each point, the coordinates from the reference simulation [48] and the coordinates from this simulation. In order to find the coordinates of point B, the distance SPP B must be calculated. Let us now consider the stage where the HAP is not performing any rotation. From Figure 5.13 and Table 5.8, the distance SPP B can be calculated using the following equation: SPP B theta = A SPP ( ) tan Equation [5.5] According to Table 5.7, theta is equal to degrees and the height of the HAP is A SPP = 17km. Therefore, the distance SPP B is found to be approximately equal to 5455m. Now, in order to deduce the coordinates of point B, the phi angle is used which in this case is equal to 60 degrees. Thus x B and y B are found to be equal to: ( 60) SPP B m x B cos = 2727 = Equation [5.6] ( 60) SPP B m y B sin = 4724 = Equation [5.7] Repeating the same calculations for all the set of angles, the following table was created: Table 5.10 Positions of virtual base-stations from reference simulation [48] V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z

122 CHAPTER 5. HAP Communication Model System Design Verification Repeating the same calculations as above for the case when the HAP rotates 10 degree with respect to its x-axis (pitch), the new position of the virtual base-station was found. Z X Y Pitch A theta z Cell B slides to position B in a hyperbolic manner. B(x p,y p ) B (x p,y p ) y SPP phi x Figure 5.14 Example of a HAP when it rotates with respects to its x-axis Since the height of the HAP remains the same, the distance SPP B is equal to: SPP B' theta = A SPP ( ) tan Equation [5.8] From Table 5.8 the theta angle after the 10-degree pitch effect became from what it was before (17.79 ) whereas the height of the HAP (ASPP) remained at 17km. Applying Equation [5.8], SPP B = 4672m. Now, in order to calculate the coordinates of point B, the φ AZIMUTH angle is used. Thus x B and y B are found to be equal to: ( ) SPPB = m x B cos 263 = Equation [5.9] 121

123 CHAPTER 5. HAP Communication Model System Design Verification ( ) SPPB m y B sin = 4665 = Equation [5.10] Repeating the same calculations for all the set of angles, the following table was created. Table 5.11 Positions of virtual base-stations from reference after 10 pitch effect V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z In order to compare the results above, which are from the reference simulation [48], with our simulation code results, it is necessary to generate the same initial positions (without the rotation) x B and y B. So, it would be expected that the resultant new positions would be the same as in Table Then the roll effect can be applied and if the results are the same as in Table 5.11, then it can be said that the two methods agree. In order to generate the same initial position (x B,y B ) using our simulation code, it is necessary to input the right size of cell radius. To present this, it is necessary to define the concept of the radius - R of a hexagon. This is defined as the radius of the largest circle enclosed by the hexagon. The simulation code set_base.m accepts the cell radius as a parameter in order to allocate the positions of the cells. In this example, there are 7 cells in the system. Therefore, all cells (except the central one) will be positioned at a distance of 2 times the cell radius. The following figure illustrates the concept of cell radius and the positioning of the cells in this simulation code. The relation between SPP B and R i is that distance SPP B = 2Ri. 122

124 CHAPTER 5. HAP Communication Model System Design Verification B SPP Re Ri Figure 5.15 Definition of internal (R i ) and external (R e ) cell radius in the Simulation Code Since distance i.e. R i = 2727m. SPP B was found to be 5455m, then the radius should be half of this distance; Running our simulation model, the position of the cells matched exactly the positions calculated previously based on the set of angles (Table 5.10). Table 5.12 Positions of virtual base-stations applying the reference simulation parameters V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z Now, it is expected that when performing the 10-degree pitch effect (with respect to the x-axis), the 3 rd virtual base-station in the example shown before will have the same position as the one calculated before (point B ). Performing the 10-degree roll effect generated the following table: 123

125 CHAPTER 5. HAP Communication Model System Design Verification Table 5.13 Positions of virtual base-stations when HAP rotates with respect to its x-axis V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 x y z To get a better view the results from our simulation and the reference simulation are presented together. The following table (Table 5.14) represents the initial position for both the reference simulation and our simulation whereas Table 5.15 represents the 10-degree roll effect of the HAP. For the initial position, the first set of x and y coordinates refers to the reference simulation results and the second set of x and y coordinates refers to our simulation results. Table 5.14 Verification of Results before the 10-degree pitch effect V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 Reference Simulation Results x y z Simulation Results x y z For 10-degree pitch effect, the first set of x and y coordinates refers to the reference simulation results and the second set of x and y coordinates refers to our simulation results. 124

126 CHAPTER 5. HAP Communication Model System Design Verification Table 5.15 Verification of Results after the 10 pitch effect V. Base station 1 V. Base station 2 V. Base station 3 V. Base station 4 V. Base station 5 V. Base station 6 V. Base station 7 Reference Simulation Results x y z Simulation Results x y z The following map, illustrates the virtual base station positions on the ground before and after the rotation took place. 6 After Before SPP y -5-6 x Figure 5.16 Pitch Effect Rotation with respect to the x-axis for both reference and current simulation (top-view) 125

127 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.17 Pitch Effect Rotation with respect to the x-axis for both reference and current simulation (side-view) Table 5.14 and Table 5.15 show that the pitch effect performed using the 3D rotation technique [46] generates the same results to those used in [48]. From Figure 5.16 and Figure 5.17 it can also be seen that the pitch performed, caused the distortion of the entire hexagonal shape of the coverage area. This not only applies on the shape of the overall coverage area but as we will see in the following chapter, it affects almost every cell. In the following section, the effect of rotation of the platform is presented in terms of received power on the ground Verification of CIR levels At this stage, it is required to verify the Carrier to Interference Ratio (CIR) levels for each receiver (user). This requires using Equation [3.21] and Equation [3.22] (azimuth and elevation angle calculations for received power on the ground) for calculating n θ and n φ and Equation [3.23] for calculating the directivity. The receivers in this model are positioned uniformly in the shape of a square grid as shown in Figure The number of receivers per square meter defines the resolution of this grid with respect to the total coverage area. The coverage area is defined as the area enclosed by the circle with diameter (d). The receivers are uniformly lined up within a square that its sides are of size (L). The size of L is set to be larger than d in order to get a full picture of the received power and CIR or the Carrier to Noise plus Interference Ratio (CNIR) levels. 126

128 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.18 CIR Testing Model - Receiver Positioning Every user has different directivity level with respect to every antenna on the HAP calculated using the Equation [3.23] presented in Chapter 3. Calculating the directivity levels, it is possible to find the CIR levels using the following equation. CIR ( x, y, z) = Nc i= 1 Power i Power MAX ( x, y, z) ( x, y, z) Power ( x, y, z) MAX Equation [5.11] where N c is the number of co-channel cells. The function which is implementing this task, searches for the highest power level for each user position (x,y,z). This is also the maximum power from an individual beam and therefore the carrier. This value is then divided by the sum of powers in all other beams (always for the same user), which is the interference. A number of simulation scenarios have been investigated. The purpose of these scenarios was to: 1. Verify the communication base part of the simulation model described in this chapter. 2. Investigate the effect of cluster size on the received power and CIR levels of the system. 3. Investigate the effect of a rotational movement such as pitch on the CIR levels of the system. 127

129 CHAPTER 5. HAP Communication Model System Design Verification All simulations scenarios performed, were using the parameters listed in Table Table 5.16 Simulation Parameters No Parameter Value Unit 1 Coverage Area Diameter (d) 60 km 2 Receiver Area Length (L) 66 km 3 Number of Bases (N B ) Cluster Size (K) 1, 4, 7 5 Platform Height (h) 17 km 6 Number of Receivers (N RX ) Cell Radius (R) 3.15 km The first step before verifying the results for the communication model was to make sure that the n θ and n φ were calculated correctly. This was possible as the mathematical model (Figure 3.15) used to calculate these two values was the same as the one used in the past for previous HeliNet models [48]. The results from the communication model were verified using the results from HeliNet report T1 [56]. These are the results from the plots of the received power contours for one antenna on the HAP. The user positions were used to monitor the received power signal level. The cluster size for this example is set to 7 and the total number of beams is 127. Sidelobes have been modelled as a flat floor at -40dB. Figure 5.19 illustrates the three different cases of cluster size (K) examined. For this example we will be focusing in cluster size of

130 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.19 Cluster size of 1, 4 and 7 Figure 5.20 illustrates the normalised to the peak received power contours for the centre cell. As you can see there is approximately a 4.5 db drop from the centre of the cell to the edge, which is set to be 3.15km long. 129

131 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.20 Centre Cell Received Power Contour We can also see in Figure 5.20 that the received power becomes -40dB at about 10km from the centre of the cell. This indicates that the cells will overlap each other if the minimum received power threshold defined for the users permits. In a real case scenario, overlapping cells will be the case and provided that they are assigned with different channels, the system will operate with limited co-channel interference. In the contrary, it can be proved that, overlapping between cells can be beneficial for the user capacity of the entire HAP communication system [42], [62]. 130

132 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.21 CIR coverage for cluster size equals 7 Figure 5.21 illustrates a set of 127 cells positioned in an area of radius of 30km. The centre-tocentre distance is 5.46km. This is because the radius (R) of the hexagonal cell illustrated above was set to be 3.15km. Recalling from section 4.2, page 84, the value of the radius used in the BaseLoc code, defined the position of the cells in a congruent way. 131

133 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.22 CIR profile across the coverage area for cluster size equals 7 Figure 5.22 illustrates the CIR profile across the coverage area for cluster size (K) of 7. From this plot, it can be seen that the CIR levels become lower for the cells closer to the centre of the coverage area. More specifically the CIR levels drop by 10dB from the cell at the edge of the coverage area to the cell on the centre of the coverage area. This is because the centre cell is surrounded by a number of interferers that contribute to the reduction of its CIR levels. Therefore, the centre cell interferes with 12 neighbouring cells (see Figure 5.19 Cluster Size 7) operating on the same channel (0). Cells located far from the centre, experience higher CIR levels since the number of interferers is expected to be lower. The angular separation when K is 7 is greater than the case where K is 4 or even 1. This means that the interference from neighbouring main lobes, which operate at the same channel, is less dominant. Using Equation [1.3] (page 33), the frequency reuse distance for K of 7 is 14.4km, for the case of K of 4 is 10.9km and for the case of K of 1 is 5.46km. In the following section, the case of K equals 4 and 1 will be examined. This is to examine how cluster size is related to the CIR levels. 132

134 CHAPTER 5. HAP Communication Model System Design Verification Cluster Size effect on CIR The cluster size (K) for this example is now set to 4 and the total number of beams is 127. Sidelobes are set to be flat at -40dB. As seen in Figure 5.23, the CIR levels for K of 4 is about 5dB lower than the case of K of 7. This justifies the above statement that the angular separation of K of 7 is greater than K of 4. This can also be seen in Figure 5.19 Cluster Size 4 since the cells using the same channel (interferers) are more than the case where K is 7. Figure 5.23 CIR coverage for cluster size equals 4 Figure 5.24 illustrates the CIR contours for a cluster size (K) of 1 and a total number of beams of 127. Sidelobes are set to be flat at -40dB. As seen in Figure 5.24, the CIR levels are much worse than the case when K is 4 or 7. The CIR drop significantly from the centre to the edge of each cell. For this case we effectively have overlap between beams that are using the same channel (see Figure 5.19 Cluster Size 1) and therefore they interfere with each other. This is an extreme case and it is not practically useful. Nevertheless these results have been useful to show the significance of sustaining low interference and therefore high CIR levels. 133

135 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.24 CIR coverage for cluster size equals 1 134

136 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.25 CIR profile across the coverage area for cluster size equals 1 Figure 5.25 illustrates the CIR profile across the coverage area for cluster size (K) of 7, 4 and 1. The CIR levels for K=4 are now lower than the case of K=7. This time the CIR increases when moving away from the centre of the coverage area and decrease after the 3 rd cell at about 20km from the centre of the coverage area. In addition, the CIR levels for K=4 have dropped by about 5dB and for K=1 by about 20dB when compared with K=7. It is also interesting to observe that when K=7 the link budget improves towards the outer cells. Effect of 10 degrees pitch on received power and CIR levels Figure 5.26 Illustrates the received power when the platform performs a 10-degree pitch movement. The centre of the cell has now moved at about 3km along the x-axis. Also the power contours on the ground are distorted. This is because the beam is circular (referring to the centre cell) and when projected with an angle on the ground other than 90 degrees, the result will be a distorted cell. Nevertheless the power roll-off from the centre of the cell to the edge (at 3.15km) remains the same. This indicates that the centre cell will still be able to cope with minimal 135

137 CHAPTER 5. HAP Communication Model System Design Verification distortion to its nominal area of coverage meaning that it will still be capable to provide coverage within a radius of 3.15km. Figure 5.26 Centre Cell Received Power Contour 10 degrees pitch This does not apply for the cells at the edge of the coverage area. Some of these cells (see Figure cells at the right hand side of the edge of the coverage area), will experience much lower CIR levels. This is primarily because the path-length for these beams is longer and also because the angle of elevation for the users is smaller. 136

138 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.27 CIR coverage for cluster size of 7-10 degrees pitch Figure 5.28 illustrates the CIR profile across the coverage area for cluster size (K) of 7, when the platform performs a 10-degree pitch movement. The CIR levels have dropped significantly at the right-hand side of the coverage area. As mentioned above this is because of the longer path length the beams at the edge experience. 137

139 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.28 CIR profile across the coverage area for cluster size of 7-10 degrees pitch The shape of the cells will become circular again only if the HAP returns to its steady state with zero pitch angle. A lot of work [48] has been done in the Communication Research Group - University of York in order to limit the effect of the rotation of the HAP on the coverage area. Technical and theoretical solutions are currently under development such as steering antenna platform in order to minimise the effect of roll or pitch of the HAP. Various techniques have been investigated and will presented in Chapter 8. In any case the aim is to ensure adequate CIR levels and most preferably uniform CIR levels in the coverage area. This is to ensure that all users connect with the same modulation scheme and therefore experience the same data rates. Furthermore to be able to guarantee that all areas of the coverage area will be provided with coverage at all times Verification of the Traffic Model At this point we present the verification of the Monte Carlo simulation model based on the ErlangB type of traffic distribution. The verification was performed on one cell to keep the model as simple as possible and to ensure that the results agreed with the theory (ErlangB formula presented in Chapter 6, Section Equation [6.7]). 138

140 CHAPTER 5. HAP Communication Model System Design Verification One-Cell model simulation Initially, one cell has been considered as the coverage area. The aim was to keep things simple to understand, to implement and verify. Then the model was extended to more cells after being designed, developed and tested. Base Station Figure 5.29 One-Cell Representation Top view HAP Base station is virtually based on the ground Figure 5.30 One-Cell Representation 3D view 139

141 CHAPTER 5. HAP Communication Model System Design Verification Figure 5.29 and Figure 5.30 depict the users positioned randomly within the circular coverage area as a top view and a three-dimensional view respectively. The base station is considered as being positioned above the centre of the cell. In terms of a HAP, the base station is located on the platform. At this point, for simplicity, we do not consider user-received power being affected from the height of the platform. The users can directly connect to the base station without worrying about their Carrier to Interference Ratio (CIR) assuming that there are available channels into the system. This forms a very basic channel allocation scheme that allows a fair comparison between the two models. In addition, the code is written in such a way that it does not allow any user to attempt to switch on and back off within the same simulation time step. The state of every user is represented by a variable called state. The user can be in the off-state (and at anytime can decide to make a phone call), the blocked-state (in which the user is blocked and is not allowed to make a phone call again unless he/she has been unblocked) and finally the going state, which represents the state where a user is talking on the phone. The following figure illustrates diagrammatically the algorithm used for the Monte Carlo model. At the beginning the code passes through each user examining its status. If the status is off (initially all users are set to be off ), the user is asked whether he/she wants to make a phone call. If the answer is no, the status of the user remains off and the code moves on to the next user. If the answer is yes, then the code examines whether there are any channels left and if there are any, then the user is allocated a channel and its status becomes going. In the case where there are no available channels, the user is blocked and its status is updated to blocked. 140

142 CHAPTER 5. HAP Communication Model System Design Verification Next User no For N users Next User no Switch off State Blocked Check Status State Going Switch off yes Channel Released yes State = Off no State Off Switch on State = Off yes Originating call arrival Channel Available no Call Blocked yes Channel Assigned State = Blocked State = Going Next User Figure 5.31 Monte Carlo Model - Algorithm Analytical Diagram If the status of the user is already blocked, then the user is asked whether he/she wants to have their status changed back to off allowing them to retry to make a phone call at a later stage. The probability of a user having its state changed from blocked to off is equal to the probability of a user having its state changed from going to off. However, as far as their state remains blocked, these users will not be able start a phone call. It is required to maintain the offered traffic at a constant rate, irrespective of the blocking probability. This in turn requires that the number of users attempting to start a new phonecall in every timeslot is independent of the blocking probability. This enables the offered traffic to be specified in advance of running the simulation. Consider the case where users, once blocked, are immediately available to attempt to make a new call. The number of users attempting to make a phonecall would then be a function of the number of users not on the phone. As the blocking probability increases, then the number of users not on the phone will increase, and for a given probability of starting a new phonecall, a larger number of phonecalls will be started. Hence the offered traffic will increase. However, if blocked users are held in a "blocked" state for the time that they otherwise would have been on the phone, then the number of users available to start a phonecall is independent of the blocking probability - since blocked or not, users attempting to 141

143 CHAPTER 5. HAP Communication Model System Design Verification start a phonecall are prevented from trying again for the same time. Hence in this case, the offered traffic is independent of the blocking probability. In the case where the user is already going, the user is asked whether he/she wants to terminate their phone call. If the answer is yes then the user status becomes off and a channel is released. If the answer is no, then the user continues to talk on the phone. The algorithm is repeated at the beginning of every frame giving the chance for all users to change their state (except those who are blocked). Verification of the One-cell simulation model In order to ensure that the queuing analysis presented before is correct, a model based on onecell has been developed. This model was used along with the theoretical model based on the ErlangB equations to verify that the simulation results were consistent with the analysis. The verification has been performed using the error bar technique [63]. The design, implementation and verification of the model was based on the following rules: 1. Ensure that the Monte Carlo Model was performing the same task (validation of the model). 2. Ensure the same parameters / environment of operation for both models. 3. Approximations and assumptions had to be made as close as possible to the theoretical model (ErlangB equations). 4. Use the error bar technique to make sure that the results lie within the confidence limits. The first step was to validate the Monte Carlo model, i.e. to make sure that the Monte Carlo model performed the same task as the theoretical ErlangB model. To do so, the general characteristics of an ErlangB model listed before (section page 94), had to be reflected in the assumption of the Monte Carlo simulation. Also, in order to perform a fair comparison between the two models, they had to be tested under the same parameters. More specifically the number of channels and the pre-specified offered traffic per square unit had to be the same. For the Monte Carlo based simulation, the number of users was set to be a large number (it is assumed infinite in the analysis). Also, the mean call length was set to be large enough to ensure the correct operation of the random generator (this is explained further in the following section). The channel allocation scheme was also very basic: a limited number of channels were available to a large number of users based on the principle of the first come first served. 142

144 CHAPTER 5. HAP Communication Model System Design Verification Generating Calls Technique and the Random Number Generator Initially the user was allowed to switch on if the value from the random number generator was smaller than the probability of a user to start a phone call (Probability Off to ON -Table 4.1). This probability was a very small number of the order of The algorithm was initially written as follows: if ( user_state == off) && ( random_number < probability_to_go) ) { } Look for available channel According to the error bar technique, the results generated using this piece of code to allow a user to switch-on, showed that the blocking probability calculated was always lower than the theoretical value based on the ErlangB model. Table 5.17 lists the parameters used for this simulation. Table 5.17 Simulation Parameters for one cell Number of Cells (c) 1 Channels per Cell 14 Offered Traffic 3erlang/sq. unit area Users 5000 Conversations 5000 Coverage Area 3.14sq.units Average Call Length 6 minutes Arrival Process Poisson The following error bar plot is generated for the Monte Carlo model based on the algorithm presented above with a confidence limit of 99%. 143

145 CHAPTER 5. HAP Communication Model System Design Verification erlang-b theoretical result Figure 5.32 Error bar results proved wrong in the first approach allowing a user to switch on Using ErlangB standard equation (Equation [6.7]), the blocking probability of a server with these parameters will be In the approach presented above the average blocking was for 100 independent runs. This is 1.87% lower blocking than the theoretical value and as seen in Figure 5.32 the theoretical value lies outside the confidence limits. The same error (lower blocking than the theoretical one) was persistent despite the fact that the number of users and the simulation time step were varied. Repeating black box and white box testing [64] on the code lead us to the conclusion that there was something wrong with the random number generator. Random Number Generator At this point it was important to investigate further the behaviour of the random number generator since many questions have been raised with respect to the general behaviour of the generator. The initial observation made about the random number generator was that when decreasing the mean call length or the total number of conversations, the code was generating exactly the same random numbers. As the seed of the random number generator is based on the computer time, in 144

146 CHAPTER 5. HAP Communication Model System Design Verification units of one second, a small mean call length or small number of conversations meant that the simulation had enough time within 1 second to run over again. As a result the overall blocking and offered traffic in the cell was exactly the same for several consecutive independent runs. It was therefore important to ensure that the routine is not called twice within the same second. In order to do this, the number of conversations during each simulation was increased. However, the problem of the Monte Carlo code consistently generating low blocking probabilities was still present. The first possibility of why the blocking was consistently lower than the theoretical value was that the random number generator was not generating enough random numbers below the probability of a user to switch-on. As mentioned before, the probability of a call to occur is very small and the value generated from the random number generator must be even smaller to allow the user to switch on. If this was the case (that the random number generator was not generating enough random numbers) then the code would allow fewer users into the system and therefore the OT should be significantly lower than the analytical prediction. To find out whether the random numbers were uniformly generated, the code was tested. At first, the offered traffic that would have caused this blocking level (Figure 5.32) was calculated and then it was compared with the nominal value set in the code. This was done by first calculating the offered traffic (OT) that would have caused the simulated level of blocking probability. Then it was found out how wrong the random generator would have to be in order to do that and the results were compared with how wrong it actually was. This allowed us to find out whether it was possible for the random number generator alone to cause this drop on the blocking probability. The OT which dependents on the uniformity of the random generator, it has always been very close to the specified value. Verifying that the random number generator is always uniform, and that the OT generated is always the right value leaves us only with one conclusion. That in fact the blocking is lower than the theoretical one because of the pattern these random numbers are generated. It has been concluded that the problem with the random number generator was not that it did not produce small number numbers but that it did not produce them together. In other words, the random number generator code tended to space them out since the random numbers that it produced were not completely uncorrelated [57]. Therefore, the calls were not independent because of the pattern in the random number generator. As a result, the code presented 145

147 CHAPTER 5. HAP Communication Model System Design Verification previously has been very sensitive not only to the number being very small but also to the pattern of the very small numbers generated. In order to fix this problem, and ensure the users operate independently, the code was modified to be less sensitive to the pattern of very small numbers generated from the random number generator. if ( user_state = = going) { if ( random_number_1 < square_root_probability_to_go) ) { if ( random_number_2 < square_root_probability_to_go) ) { Look for available channel } } } The way that this code works is that the random number generator code will be called twice and each time is checked to see whether the number generated is below the square root to the probability to go (i.e. probability to start). That makes the total probability the same but it s not reliant on very small numbers anymore. The following error bar plot is generated for the Monte Carlo model based on the algorithm presented above with confidence limits of 99% and the simulation parameters listed in Table

148 CHAPTER 5. HAP Communication Model System Design Verification erlangb theoretical results Figure 5.33 New approach on how to allow users to switch-on was successfully verified using the error bar technique The average blocking probability in this case was Figure 5.33 shows that this value lies within the error bars. This plot gave us the confidence to continue the development of the Monte Carlo model based on the algorithm presented above Conclusions In this chapter we have verified the correct operation of the HAP simulation model presented in Chapter 4. A systematic testing mechanism was followed in order to ensure complete system analysis and verification of the simulation model. From the results it has been concluded that a HAP system can deliver a symmetrical hexagonal cellular layout. Changes in the clustering or directivity of the antennas has a direct impact on the size and shape of the cells on the ground as being defined from their boundaries (minimum received power levels Chapter 3 Section 3.6). Also, the movements of the HAP are causing significant changes to the received power levels at the user s end on the ground that is significantly affecting the coverage in some regions. 147

149 CHAPTER 5. HAP Communication Model System Design Verification The advantage of the symmetrical cellular layout of the HAP on the ground is being exploited in great extend and it is presented in the following chapter. This results from the importance of the capacity in the system and trying to find ways to maximise it. As will be seen in the following chapter, the size of each of the cells is symmetrically increased allowing neighbouring cells to overlap with each other forming areas of overlap. This overlap is being exploited in order to increase the overall system capacity. The traffic model designed and implemented was verified with the theoretical values. The average blocking probability was found to be lying within the error bars giving us the confidence to continue with the development of the Channel Allocation schemes as well as the Handoff schemes based on this simulation platform. 148

150 CHAPTER 6. Fixed Channel Allocation Based Techniques Chapter 6. Fixed Channel Allocation Based Techniques 6.1 Introduction Fixed Channel Allocation Based Techniques Cell Overlap Fixed Channel Allocation based Schemes Numerical Calculation for Uniform Blocking (NCUB) Sensitivity to numbers of channels per area Summary and Conclusions Introduction Fixed Channel Allocation (FCA) scheme is a simple channel allocation scheme that works well with regular hexagonal cell structures based on a cellular architecture when there is no shadowing. HAPs combine all these characteristics and in conjunction with their physical properties it has been possible to develop new channel allocation schemes that are based on FCA. The purpose of this chapter is to look at the effects of different numbers of channels within each of the different regions and to look at the effect of the granularity of the number of channels available rather than actually generate specific performance results for a scheme with practical applicability. This chapter begins with the description of a basic FCA scheme. Following, the concept of cell overlap is introduced and analysed. Based on the cell overlap, a number of more sophisticated FCA-based schemes is been investigated. From these schemes it is shown that the cell overlap improves the QoS in terms of the overall blocking probability. This is because of the higher trunking efficiency that the overlap areas experience. Nevertheless, the blocking across the different types of areas of overlap formed is uneven and therefore not fair. To try and maintain a fair service (uniform blocking), a channel allocation technique is proposed that takes into account the number of channels available, the offered traffic (OT) and the cell overlap in order to perform the channel allocation. This scheme improves uniformity in terms of blocking but its performance depends on the number of channels available per cell. The distribution of the channels in a cell is further analysed to investigate the sensitivity of the system in terms of blocking probability for different levels of trunking efficiency. Finally, the scheme that achieves uniform blocking is examined under different system parameters such as different size of overlap or different OT. 149

151 CHAPTER 6. Fixed Channel Allocation Based Techniques 6.2. Fixed Channel Allocation (FCA) Scheme The main characteristic of the FCA schemes is that they divide the channel allocation amongst the cells giving a specific list of channels to a particular cell (see Chapter 2 Section 2.1). Therefore the number of channels available in each cell depends on the number of cells in a cluster. A cluster is generated assuming non-overlapping uniform hexagonal cells as shown in Figure 6.1. Frequency reuse factor also know as cluster size (K) takes place by tesselating the clusters - assuming that the network is made up of a uniform hexagonal grid of base stations. Figure 6.1 illustrates the case where K is equal to 3, 4 and 7. As mentioned before, the concept of hexagonal cells is based on the fact that all base stations are positioned based on a hexagonal grid and the users connect to the closest base station. The results presented in this chapter have been generated with cluster sizes of 7, with a fixed frequency allocation shared equally amongst the base stations. This is to exclude interference from the initial results since the total number of cells examined was 7. Having more cells in the cluster means that: The frequency reuse distance is effectively increased. This reduces the level of interference (I) and increases the carrier to interference ratio (CIR) at both the mobile and base station. However, the trunking efficiency is reduced since each base station will now have fewer channels available. Trunking inefficiency has become apparent to regulators [65] who are trying to increase competition in the cellular market by allowing more companies to compete. The frequency allocation is performed based on the number of operators that want to share the frequency spectrum. Therefore, each operator gets a small amount of bandwidth (approved by Government s National Body e.g. OfCom UK). Each provider might then have to subdivide these channels into smaller groups around the cells based on the channel allocation scheme and the cluster size that is employed. The division of the spectrum into smaller groups of channels results in poorer trunking efficiency because each company controls fewer channels per cell. However, in the case of the HAP communication systems, when the operator employs more than one HAP, it is possible for its users to switch between the platforms. By doing this and assuming that the users employ 150

152 CHAPTER 6. Fixed Channel Allocation Based Techniques directive steerable antennas, it is possible to reuse the same channels in the same area by allowing spatial diversity to take place. As a result the trunking efficiency can be improved [42] Cluster Size: 3 Cluster Size: Cluster Size: 7 Figure 6.1 Examples of cluster size arrangements In the case of HAP communication systems it is not yet clear how these systems will be practically operated. This depends primarily on the number of operators and the number of platforms available that provide services to the coverage area. To be more precise, a list of possible scenarios is presented below: 1. The simplest scenario is when one operator is using one platform. In this case the operator can have all the channels available for its users. 2. In the case where two or more operators share the same platform to serve their users located within the same coverage area, then the channels will be divided between the providers. As a result, the trunking efficiency will be reduced. 3. If the service demand in the coverage area is high, then more platforms can be brought in and each operator might use a different platform to serve the same coverage area. This is practically feasible, as HAPs will be positioned far from each other meaning that users will be pointing their directive antennas into different directions in the sky. In this case we have a multiple HAP constellation [42] [66] where the service area of each platform overlaps with the others. The users located within this area must choose the right platform to point according to their service provider. Due to the increase in the number of platforms, the number of available channels will also increase and so more users can be supported. 4. A more complex scenario is when multiple operators use multiple HAPs. I.e. each operator can use more than one platform at a time depending on the number of its 151

153 CHAPTER 6. Fixed Channel Allocation Based Techniques customers. Like the previous scenario, the service area of each HAP will overlap with the rest but this time the users are allowed to connect to any of the platforms regardless of who might be their provider. Thus the network becomes less prone to failure since when a platform fails to work for any reason the rest can serve the coverage area and the users can connect to any of the remaining ones. For this part of the system analysis, we have assumed that a single operator will be providing services to a nominal coverage area. Issues such as trunking efficiency and frequency reuse distance will still impose performance limitations and will therefore be addressed later in this chapter Cell Overlap As mentioned above, the cells formed on the ground are assumed to be circular and of equal size; this can be readily achieved by a careful design of the antenna beam profiles [50]. Interference between cells is largely due to the gain profile and sidelobe levels of the antennas used. It has been shown that the antenna gain profiles at the cell edges can create useable cell overlap [43]. Although cell overlap does occur, it has not always been used to redirect traffic from one cell to another. One example is the case of fixed wireless terrestrial systems where the user antenna requires to be redirected when moving between cells, a comparatively slow process, and during this time the user will be out of contact with any base station. In addition, this will require the system to be highly centralised, due to the need to exchange data about the channels available within each cell at high speed [25]. Nevertheless the concept of overlapping cells has been investigated in the past [40] [67] for mobile terrestrial systems with schemes such as directed retry (DR), directed handoff (DH) and a variety of selective handover for traffic balance (SHOT) schemes. From work [40] it has been shown that with the DR, an increase in the overlapping between cells leads to an increase in the quality of service (QoS) provided by the system. Furthermore, the DH scheme proved to have good sensitivity properties with respect to variation in the spatial profile of the system. From work [67] it has been shown that the SHOT schemes improve traffic handling capacity and enhance frequency utilisation. The more cell overlapping the more the traffic carried provided that the interference is acceptable. Most of the practical problems that the terrestrial systems face are not applicable in the case of HAPs. This is because of the nature of the system: all transceivers are co-located on the platform, and the platform provides a line of sight communication link with the stations on the 152

154 CHAPTER 6. Fixed Channel Allocation Based Techniques ground. This means that there will be fewer obstacles between the users and the platform (therefore no shadowing and multipath when operating in the mm-wave band), and the cell overlap can effectively be applied in all cells of the system. The HAP itself can therefore keep track of all channels in use within its coverage area by making use of a centralised architecture. The footprints are positioned to cover a nominal coverage area. Their size must be large enough not to leave any parts of the coverage area unattended. An example can be seen in Figure Figure 6.2 Illustration of a typical 7-cell (beam) scenario The overlap occurs when the sidelobes formed by the transmitting antenna, transmit signals greater than the minimum power threshold. The size of the overlapping area can be determined by setting a minimum received power limit or Channel to Interference Ratio (CIR). This is determined from the link budget and Channel to Interference Ratio (CIR) values. Calculation of the link budget is related to the power roll-off of the antenna plot. The fact that the cells overlap with each other can be proved beneficial for the performance of the system since this allows a better allocation of channels to the cells. As mentioned before (Chapter 3, Section 3.1), users can chose a base station based on their distance from that station or based on a minimum received power threshold. In either case, cells are set to overlap with each other as shown in Figure 6.2 to ensure full coverage. However, for the case where users connect to the closest base station the cell overlap is effectively ignored since despite the extent of the cell coverage, users will still connect to their closest base station. To take advantage of the cell overlap users must be allowed to choose to connect to any base station that happens to be within range (i.e. have the minimum received power threshold or minimum CIR required). 153

155 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.3 illustrates the concept of the cell radius. Here, R i defines the radius of the circle enclosed inside the hexagonal cell (internal circle) and R e the radius of the circle which encloses the hexagonal cell (external circle). R i R e Figure 6.3 Cells with no overlap The value of the overlap radius R (see Figure 6.4) varies within limits: the minimum value of the radius of the overlapping cell is equal to the original radius of the cell R e in order to avoid leaving any areas without service (if R was less than R e ); and the maximum value is taken here to be equal to 1.5R e radius, in order to limit the maximum number of overlapping cells to three. The aim is to limit overlap to only three cells instead of four or even more and to prevent the system from becoming overly complicated by numerous overlapping cells, which would result in co-channel interference. In theory, four or even more cells can overlap if the cell radius is increased sufficiently, assuming the co-channel interference and received power levels remain acceptable. 154

156 CHAPTER 6. Fixed Channel Allocation Based Techniques Ri Re R Figure 6.4 Cells considering overlap Figure 6.4 depicts the case when overlap radius R becomes greater than the initial external cell 2R i radius R e or internal radius R i. The relationship between R e and R i is R e=. The reason for 3 presenting internal radius R i is because the mathematical solution was based on it and therefore it is used in all the equations defining the overlap areas. R e is however used to define the limits of overlap and plot the graphs since it implies that there will not be any areas without service. The overlap radius R will be normalised either with respect to the external circle radius R e or to the internal circle radius R i (depending on the simulation). The radius (R) of the overlapping circles can extend up to the point where R max = 1.5R e or Rmax = 3R i. This is the maximum point for which overlap is restricted to being between just three cells. Table 6.1 Limits of Overlap for the communication model Radius of Overlapping Cell R Min R Max Limits Equal to R e. This is the minimum value of radius so that all the coverage area will be served. Equal to 1.5R e. In order to avoid overlapping between 4 or more cells. 155

157 CHAPTER 6. Fixed Channel Allocation Based Techniques Cell - 1 Cell - 2 Re A1 B1 C A2 R B3 B2 A3 Cell - 3 Figure 6.5 Illustration of Regions 3 cell example Whenever two or more adjacent cells overlap, they form a set of individual regions, which can be categorised into three types (A, B or C) according to the numbers of cells that overlap at any given time. They can therefore be assigned a channel from one, two or three cells respectively. If we sum up the regions in one cell according to the three types of degrees of overlap, we then have what we call areas. To give a more complete picture of what a region is and what an area is let us assume that seven cells overlap each other like the case in Figure

158 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.6 Illustration of Regions - 7 cell examples In this case the centre cell is partitioned into six regions of (overlap) type B (B1to B6), six regions of type C (C1 to C6) and one region of type A. If the regions (of the same type) illustrated in Figure 6.6 are grouped together, they will form three types of areas (again of type A, B and C). Figure 6.7 illustrates these three types of areas. In the seven-cell overlap example, the size of Area A equals to the size of region A, the size of area B equals to the sum of regions B1 to B6 and the size of area C equals to the sum of region C1 to C6. 157

159 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.7 Illustration of Areas - 7 cell examples The significance of having regions and areas is that it is possible to perform the channel allocation based on either region or area, which will be shown later in this chapter. If for example the channel allocation is based on regions (assuming a fixed channel allocation scheme and a uniform offered traffic), then each region will be assigned a nominal number of channels. The number of channels depends on the size of the region and the assigned channels can only be used in that specific region. If on the other hand the channel allocation is based on areas (i.e. a sum of regions), then the channels allocation should be performed according to the size of the area. Therefore, the total number of channels allocated to an area will be greater than the number of channels allocated to a region. These channels can be assigned anywhere within the area, which means that they can geographically be assigned to any of the regions consisting the area. In a general fixed channel allocation scheme, areas can be assigned different numbers of channels C A, C B, and C C. This allows a channel allocation scheme to be implemented while minimising the blocking probability P(Block). The task is to decide how to assign different numbers of channels to the respective regions, areas or cells so that the blocking probability will be the same in every part of the cell. At this point it is not directly obvious how the channels 158

160 CHAPTER 6. Fixed Channel Allocation Based Techniques must be distributed to the areas to achieve this result. However, all three cases have been considered and will be presented in detail in chapter 6 and 7. The mathematical derivation of the cell overlap is fully presented in Appendix C. The example presented in Appendix C is for three cells. Based on the equations derived in Appendix C -Table C.1, we can calculate the area as a function of the varying radius of the overlapping circles. This task requires finding the percentage that these areas cover with respect to the total area of the cell. Again, it was assumed that all three circular cells are of the same size and as a result the extended overlapping circles will also be of the same area. However, if it is required to have more than three cells overlapping the same area then the frequency reuse factor has to be greater than 3. This is because co-channel interference can occur when the size of the overlapping circles extends more than the half of the maximum frequency reuse distance (D/2) in order to include a fourth circle into the overlapping area. Although more than three overlapping cells is a feasible case scenario, it will not be attempted to be analysed due to its complexity. While varying the radius of the overlapping circles (normalised to internal circle radius R i ), the percentages of area A, B and C with respect to the size of the overlapping circle were plotted. Figure 6.8 depicts these variations. The overlapping area is extended up to 10x the initial radius (R i ) of the cell. It is however necessary to remind the reader that this is for the case where there are only three cells in the system. So, in this example it is not possible to have areas formed when more than three cells overlap each other simply because there are only three cells in the system. When more overlapping cells are considered, the maximum overlap radius will be restricted to cells. R = 1.5R = 3R simply to avoid areas formed by 4 or more overlapping max e i 159

161 CHAPTER 6. Fixed Channel Allocation Based Techniques Area A Area C Area B Figure 6.8 Percentages of Overlapping Area A, B and C with respect to the normalised radius R i - Circular shape footprints are considered for a 3 cell case scenario Before commenting the results above, it is necessary to point out that for this plot, R is normalised with respect to the internal circle radius R i. From the plot in Figure 6.8, it can be seen that when 1 R 2 area A (non overlapping area) is significantly larger than area A and C. Area B however increases almost linearly up to the point where radius R is 1.75 times the original radius R i. Also, area C remains zero up to the point where R = 2/ 3. For R 2, area A decreases significantly whereas area B starts decreasing and area C increases linearly. There is a point however, where all areas are approximately the same and this is at about 2.55 times the internal circle radius R i. This is however outside the range set previously ( R = 3R ) and it will not be considered in a real system. max i Knowledge of the behaviour of overlapping areas can be useful for the channel assignment strategy, and it is shown later in this chapter that the channel assignment will depend on these areas according to which frequency reuse factor K the system is operating at. Multi-Cell System When the cells overlapped each other symmetrically, then this formed regions A, B and C as seen in Figure 6.5. The three-cell approach (Appendix C) however is not the most common case 160

162 CHAPTER 6. Fixed Channel Allocation Based Techniques seen in a multicell system. Hence the three-cell approach was further analysed to apply to a multicell system. The following picture depicts a 121-cell HAP communication system. 5 Neighbours 3 Neighbours 4 Neighbours 6 Neighbours Figure 6.9 Circular margin of a 121 cells coverage area The coverage area is defined as the area enclosed within the bold circle. The cells marked with the number of neighbours represent the cases that need to be considered when calculating the overlap areas. The case of 6 neighbours however will be mostly considered when describing the channel allocation schemes, since it is the most common and generally applicable case. Most cells inside the coverage area have 6 neighbouring cells, which overlap with each other. According to the definition of region and areas presented before, this means that each cell (with 6 neighbouring cells) consists of 6B Small (= area B) and 6C (= area C) regions plus one region A (= area A) as shown in Figure Notice that there are no regions formed by 4 or more overlapping cells. Also note that we now have more than one type of region B (region B Small and B Large see Figure 6.12). So, from now on area A, B and C for every cell will be expressed in terms of region A, B Small, B Large and region C. 161

163 CHAPTER 6. Fixed Channel Allocation Based Techniques Region A Region B-Small Region C Figure 6.10 Inner cell consists of 6B s, 6C s and one region A Varying the radius of the cells (normalised to the internal cell radius R i ), it is observed that the three types of areas (sum of all regions) vary differently. Area A, B and C plots are depicted in the following figure as a function of the overlapping radius R. Area A Area B Area C Normalised to Ri Figure 6.11 Inner cell consists of 6B s, 6C s and one area A In the outer ring however, the cells have 5 or 4 or 3 neighbours. In the case of 5 neighbours the cell consists of 4C regions plus region B and region A. Area B this time is larger, made up of 3B Small regions plus 2B Large regions. So, it could be defined as: 162

164 CHAPTER 6. Fixed Channel Allocation Based Techniques Area B 5-Overlap = 3Region B Small + 2 Region B Large Equation [6.1] Now, area A is defined as: Area A 5-Overlap = total_cell_area (Area B 5-Overlap + Area C 5-Overlap ) Equation [6.2] Region A Region B-Large Region C Region B-Small Figure 6.12 Outer cell overlaps with 5 cells Again, varying the radius of the cells, it is observed that the areas vary differently. Area A, B and C is depicted in the following figure as a function of overlapping radius R. 163

165 CHAPTER 6. Fixed Channel Allocation Based Techniques Area A Area B Area C Normalised to Ri Figure 6.13 Inner cell consists of 3BSmall, 2BLarge, 4C and one area A Apart from the case where the outer cell overlaps with 5 cells, there is the case of 4 neighbouring cells. This is shown in Figure Region B - Large Region C Region A Region B - Small Figure 6.14 Outer cell overlaps with 4 cells 164

166 CHAPTER 6. Fixed Channel Allocation Based Techniques Here, area C is reduced to 3C regions. Furthermore, area B is also decreased and comprises two large B regions (B Large ) and two small (B Small ). Area B 4-Overlap = 2 Region B Small + 2 Region B Large Equation [6.3] As before, area A can be expressed as: Area A 4-Overlap = total_cell_area (Area B 4-Overlap + Area C 4-Overlap ) Equation [6.4] Varying the radius of the cells, it is observed that the areas vary differently. The total area A, B and C is depicted in the following figure as a function of radius R. Area A Area B Area C Normalised to Ri Figure 6.15 Inner cell consists of 2B Small, 2B Large, 3C and one area A The final case is when a cell has only three neighbours. In this case area A becomes even bigger and area B and C gets smaller; Figure 6.16 depicts this case. 165

167 CHAPTER 6. Fixed Channel Allocation Based Techniques Region A Region B - Large Region C Region B - Small Figure 6.16 Outer cell overlaps with 3 cells In this case, the total area C decreases to 2C regions. Area B is also decreased and now it consists of two large B regions and one small B region. Total B 3-Overlap = Region B Small + 2 Region B Large Equation [6.5] As before, area A can be expressed as: Total Area A 3-Overlap = total_cell_area (Total B 3-Overlap + Total C 3-Overlap ) Equation [6.6] Table 6.2 and Table 6.3 represent all the cases of overlap, with exact derivations of each area in terms of the overlap cell radius R, the internal and external normal size cell radius R i and R e respectively and θ. Table 6.2 List of Areas for various Overlapping Cases Category Area A Area B Area C Internal Cell 6 - Neighbours Cell _Area (Area B + Area C) 6 Region B Small 6 Region C External Cell 5 - Neighbours Cell _Area (Area B+ Area C) 2 Region B Large + 3 Region B Small 4 Region C 4 - Neighbours Cell _Area (Area B+ Area C) 2 Region B Large + 2 Region B Small 3 Region C 3 - Neighbours Cell _Area (Area B+ Area C) 2 Region B Large + 1 Region B Small 2 Region C 166

168 CHAPTER 6. Fixed Channel Allocation Based Techniques 2 Region A 2πR - (Area B + Area C) Region B Small Region B Large Ri 2 a cos Ri R Ri R Table 6.3 Definitions of Regions R cos( θ C ) R 2 [ ( ) + 3 ( θ sin( θ ))] [ ( ) ( ( ))] 2 2 R i 2 2 R a cos R i R Ri R 3 1 cos( θ C ) + 3 θ c sin θ c R 2 2 R 2 [ ] Region C 3 ( 1 cos( θ )) + 3 ( θ sin( θ )) where R = e R 2 i 3 C or c 3R R e i= and 2 c θ C 3 arccos 1 R c R = 2 2 2R 2 i c R i 3 2 The investigation above was useful at a later stage when the traffic was considered. It was required though, to categorise each cell based on its number of neighbours in order to obtain an accurate approximation of the total size of areas A, B and C. Since some of the cells of the outer ring (external cells) fell outside the coverage area (see Figure 6.9), where not taken into account for these calculations. In fact we have considered the case of the 6 neighbouring cells which is the most common case to be found in a large system such as the example of 121-cell system proposed in HeliNet [9] Fixed Channel Allocation based Schemes Fixed Channel Allocation (FCA) scheme is a simple channel allocation scheme that has been used for the purpose of exploiting cell overlap. FCA was used in all simulations performed, so the number of channels allocated per cell was always fixed. However, the channels allocated on a regional basis were varied. This was possible since the overlap areas could be assigned different numbers of channels CH A, CH B, and CH C. The task was to decide how to assign different numbers of channels to the respective areas, so that the blocking probability will be the same in every part of the cell. In spite of decreasing the trunking efficiency by dividing up the available channels into each area, this sub-optimal strategy has been investigated in order to achieve uniform blocking levels across all area forming a cell. The issue here was not to achieve an average minimum blocking but a minimum uniform blocking taking advantage of the overlap. 167

169 CHAPTER 6. Fixed Channel Allocation Based Techniques Problem Analysis Areas A, B and C formed in a cell, can be assigned different numbers of channels CH N-area. This will allow a robust channel allocation scheme to be implemented while minimising blocking probability P(B). The task now is how to assign the different numbers of channels CH N-area to areas A, B and C, so that the blocking probability P(B) will be the same in every part of the cell. It is not directly obvious how the channels must be distributed to the areas. The blocking probability at any point P (B) can be expressed as a function of offered traffic (OT) and total number of channels CH N-Total. The equation that describes the ErlangB distribution [58] is presented below: CH OT P ( B) = f ( OT,CH ) = CH! CH k Equation [6.7] OT k! k = 0 The number of channels (CH) distributed are dependent on the size of the three areas. For the case where we are considering one cell, the blocking probability in terms of an ErlangB distribution can be expressed as a function of the number of channels available and the Offered Traffic (OT) in the cell. P (B) = f (OT N-Total, CH N-Total ) Equation [6.8] In the case where we have more than one cell forming the coverage area and these cells overlap, then the blocking probability of a cell can be expressed in terms of the overlap. The blocking probability of a cell must now be calculated based on the fact that users are capable of choosing a channel from one cell in the case of a user in area A, or two cells in the case of a user in area B or from three cells for area C. If we categorise users based on the area they are located, it is possible to express the blocking probability of a cell in terms of its areas. The first relation to be considered is between areas and the radii of the overlapping circles, which define the percentage of overlap. This will require a mathematical - geometrical approach to the problem. Areas(A, B, C) = f (R, Re) Equation [6.9] 168

170 CHAPTER 6. Fixed Channel Allocation Based Techniques Using Equation [6.9], areas A, B and C (remember that areas A, B and C denote the sum of regions of type A, B and C respectively) can be assigned different numbers of channels based on their size. The sum of all the channels assigned per cell must however be equal to the total number of channels. CH N-Total = CH A + CH B + CH C Equation [6.10] Based on the numbers of the channels allocated in these areas (as shown above CH A, CH B, CH C ), it is possible to define the offered traffic for each of the areas assuming a constant offered traffic (k - erlang per cell, E/cell) in the cell. For area A the offered traffic can be expressed as ka whereas for area B the offered traffic can be expressed as kb and for area C as kc. The value of the areas A, B and C are expressed in percentages with respect to the total size of the cell rather than based on their actual size. Offered Traffic A = k A Equation [6.11] Offered Traffic B = k B Equation [6.12] Offered Traffic C = k C Equation [6.13] One thing to be noticed here is that the offered traffic is directly proportional to the size of the area. Combining the generalised blocking probability equation (Equation [6.8]) with the offered traffic equations; the blocking probability for each area can be expressed individually as follows: P (B) A = f (OT A, CH A ) Equation [6.14] P (B) B = f (OT B, CH B ) Equation [6.15] P(B) C = f (OT C, CH C ) Equation [6.16] 169

171 CHAPTER 6. Fixed Channel Allocation Based Techniques The equations above are valid when the users positioned in area B and C (see Figure 6.5) can only be assigned a channel from one cell. Since area B is formed when two circles overlap together and area C when three circles overlap together, the blocking probability differs from the equations listed above. Suppose that the user can chose a channel from either two cells in the case of area B or from any of the three cells for area C, the effect on the blocking probability can be seen in the following equations. P (B) A = f (OT A, CH A ) Equation [6.17] P (B) B = f (OT B,2 CH B ) Equation [6.18] P(B) C = f (OT C, 3CH C ) Equation [6.19] From the above equations it is shown that we can actually have more than 100% channels available within a cell. In essence, capacity is pulled from the neighbouring cells to the cell of interest improving the channel availability within that cell. However, it is not clear at this stage how many channels are coming into the cell of interest and which exact region within the area they are allocated to. In order give a clear picture on the channel allocation, the following three cases of channel allocation have been expressed. These are applicable when considering cell overlap. We assume CH A, CH B and CH C channels nominally assigned from each cell to each region A, B and C respectively. 1. Channel allocation on a regional basis with regional restrictions. Here, the system only allows channels allocated to a region to be used within the regions of overlap, and it restricts these channels from being exchanged between the regions of overlap. This case offers very limited flexibility since only CH A, 2CH B and 3CH C channels are available within the respective type of region. Furthermore, it has low trunking efficiency since the channels are divided amongst the regions (more on trunking efficiency can be found in Section 6.6). 170

172 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.17 Case - 1: Regional Based Channel Allocation Figure 6.17 illustrates the channel distribution amongst the regions. Users in region C (three cell overlap) will have 3 times the number of channels initially allocated to that region by each cell. Therefore the total number of channels available to the region will become 3CH C. The same applies for region B in which the total number of channels that a user will be choosing from will be 2CH B. Channel movement between each region is not allowed. 2. Channel allocation based on areas with controlled number of channels. In this case channels can be used within an area (i.e. between regions of the same type) but with some certain restrictions. For example in Figure 6.10, area B can have a total of 12CH B channels, and area C can have a total of 18CH C channels in any one cell. These channels can be deployed anywhere within the overlap area of the cell of interest but not be redeployed outside the single cell area. Figure 6.18 illustrates an example for case-2 where all 18CH C channels dedicated to area C are redirected to one of the regions C (part of area C of the cell of interest). This is a fixed limit that is being set to prevent any more channels coming into the cell. So, in the case where a user arrives in another part of area C (of the cell of interest), when there are already 18 channels across area C (as depicted in Figure 6.18), that user will be blocked although there might be channels available from a different cell. 171

173 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.18 Case -2: Area Based Channel Allocation with certain restrictions 3. Channel Allocation based on areas but this time allowing channels to move between all regions of the cell of interest. Using the example of a centre cell illustrated in Figure 6.10, channels can be allowed anywhere within the area they have been allocated to. So, channels from all regions of a cell can be directed to one region. In this example, each neighbouring cell, can redirect all its 6CH B and 6CH C channels to the common areas with the centre cell (which is the cell of interest). As a result, the centre cell can have a total of 42CH B channels allocated in area B and a total of 42CH C channels in area C. Figure 6.19 illustrates the case where channels can move around the regions (i.e. within the area they belong to). This means that a total of 18CH C channels are moved from three cells to one of their regions to give a peak value of 12CH B and 18CH C channels. 172

174 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.19 Case -3: Area Based Channel Allocation allowing channels to move between all regions of the cell of interest All three cases enabled us to look at various schemes and investigate effects such as the granularity of the channels and the channel distribution amongst the regions or areas. The main issue arising from these different cases is the fact that the number of channels available in a given cell can be uneven, as in case 2 and case 3. That makes an approximation of the ErlangB formula, but this is valid because of on the way it is used later on in this chapter. This is to look at the effect of the channel granularity and channel distribution which can be obtained from the approximation. For the scenarios where the 2 nd case is used, a centre cell (Figure 6.10) has been assumed and any overlap region in this cell should only pull 2CH B or 3CH C channels. Therefore at any time the total number of channels in the case of area B should be 12CH B and area C should be 18CH C. In this case channels are not allowed to be moved across the regions. However if we allow channels to move freely around in the area (3 rd case) it creates more flexibility but it also means that the number of channels available per unit area is spatially varying (e.g. all channels CH B used in one region). This is therefore an extreme case of flexibility where area B for example has 42CH B channels. This will benefit a hotspot, but it may adversely impact a large area of arriving users in the future. In essence we serve the hotspot at that time instant but any users arriving in the future in any part of the other regions B in these cells (with no channels left) will be blocked. The question here is should we allow this extreme case to happen and 173

175 CHAPTER 6. Fixed Channel Allocation Based Techniques serve a local hotspot? This takes flexibility in one extreme. On the other hand the first case that divides the channels into regions rather than areas will have much worse trunking efficiency and the blocking is therefore expected to be higher. This is takes flexibility into the other extreme that does not allow flexible usage of the resources. For case 2 where we have a restricted flow of channels, it has been possible to examine the channel distribution and channel granularity in the regions / areas. The aim is to maintain the same and also lowest possible blocking probability in each and every area. This can be feasible only having assigned the right number of channels to each of these areas. Analytical derivation is impractical and beyond the scope of this work, so the next best way to do this is empirically, by generating a number of results for blocking probability while varying the radius of the overlapping cells and as a result the sizes of the areas. One thing to note is that the percentages of the regions formed within a cell with respect to the cell area do not change while the height of the High Altitude Platform (HAP) changes. It is the radius of the overlapping cells that can change these percentages. Both factors will be investigated at a later stage. Nevertheless, it is expected that when the height of the HAP increases the blocking probability will also increase, and when the height of the HAP decreases the blocking probability will increase again but only in the outermost cells. For the first case where the height of the HAP increases, the total coverage area increases as well. Since there is an equal number of users per square unit area, there will be more traffic in each cell resulting on the increase of blocking. On the other hand, when the height of the HAP decreases, a percentage of the users will be out of range and hence won t be served. Therefore, the blocking probability may increase. Simulation Approach Having presented the mathematical analysis of the cell overlap, the HAP communication system traffic model will now be presented. Recalling Equation [6.17], Equation [6.18] and Equation [6.19], it can be seen that the blocking probabilities for areas A, B and C are a function of the offered traffic and the numbers of channels assigned in those areas. For calculating the blocking of the system, the ErlangB formula has been used to calculate the blocking probability. However, as mentioned before for case-2 and case-3 this is an approximation of the ErlangB distribution, which has been used to look at the effect of the channel granularity and channel distribution which can be obtained from the approximation. 174

176 CHAPTER 6. Fixed Channel Allocation Based Techniques A number of assumptions had to be made prior to the use of the ErlangB formula (Equation [6.7]): Infinite number of potential users Uniformly distributed calls with constant average call arrival rate per unit area. Blocked users never try to get back into the system after they have been blocked. As mentioned before, areas B and C are the overlapping areas where the number of channels assigned in these areas will be doubled (in the case of area B) and tripled (in the case of area C) due to the overlap. Therefore the users within these two areas will be able to choose from the channels offered from any overlapping cell, which is forming this area. As a result, the blocking probability will decrease significantly. Apart from the assumptions made for using an ErlangB distribution, we have also assumed two cases for allocating channels to the areas formed due to the overlap. The first one doesn t consider the overlapping when assigning the channels to area A, B and C whereas the second case does. What this means is that in the first case, the overlapping areas will benefit from the overlap since the number of channels in these regions will increase. For the second case, the number of channels allocated in the overlap areas is reduced so that the area with no overlap (area A) will benefit from the overlap as well. These two simple assumptions were considered to be a good start implementing a simple fixed channel allocation (FCA) model. Therefore, some of the simulations performed were based on the first and some on the second. Assumption - 1 Channel Allocation with Cell Overlap Support - (CACOS) The number of channels allocated in all areas is effectively based on the size of the individual area. In this example a centre cell has been assumed that its area B is consisted of 6B regions and its area C is consisted of 6C regions (see Figure 6.6 and Figure 6.7). CACOS is based on the 2 nd case mentioned in Section 6.4 page 170. As mentioned before, because of the overlap the number of channels in the regions B and C will be doubled and tripled respectively. The assumption is described in Equation [6.17], Equation [6.18] and Equation [6.19]. P (B) A = f (OT A, CH A ) Equation [6.17] P (B) B = f (OT B,2 CH B ) Equation [6.18] 175

177 CHAPTER 6. Fixed Channel Allocation Based Techniques P(B) C = f (OT C, 3CH C ) Equation [6.19] As a result, the total number of channels in the overlapping areas will be 2CH B and 3CH C (when the number of channels added with the channels of all the cells overlapping). Channel Allocation with Cell Overlap Support (CACOS) scheme is being used to quantify the effect of cell overlap, which had on the system performance in terms of blocking. The channels allocated in every area are based on the actual size of the area. Since we cannot have non-integer channels (but we can have non-integer areas), the channels allocated were floored. So, if the number of channels accounted in area B were 23.6, the actual number allocated would have been 23. The same applies for area C. However, area A will take the rest of the channels left in the channel pool so that no channels are left unused. Assumption - 2 Non Overlap Area Support - (NOAS) As will be seen later in this chapter, CACOS scheme improved the blocking level significantly in the overlap regions. Nevertheless, the areas left with no overlap (area A) suffer from much higher blocking levels. The Non Overlap Area Support - NOAS scheme is taking into account the cell overlap beforehand. Thus the number of channels assigned in each area is divided by the number of cells forming that area due to overlapping. So, the number of channels assigned in area B are CH B CH 2 channels and for area C, C 3. As a result, the total number of channels in the overlapping areas are now CH B and CH C (when the number of channels added with the channels of the cells overlapping). As a result the number of channels in area A will increase significantly. Like in CACOS scheme, the channels allocated in all areas are based on the size of the area. However after dividing CH B and CH C by 2 and 3 respectively, the values of CH B and CH C are floored because of not always been integers. Initial Equation CH T = CH A + CH B + CH C Equation [6.20] Modified Equation CH T = CH A ' + CH B '+ CH C ' Equation [6.21] where, 176

178 CHAPTER 6. Fixed Channel Allocation Based Techniques CHB CHB ' = 2 Equation [6.22] CHC CHC ' = 3 Equation [6.23] CH CH A ' = CH T (CH B '+ CH C ') = CH T ( B CH 2 + C 3 ) Equation [6.24] It must however be pointed out that this is not the optimum solution for allocating channels. It was considered to be useful to record the behaviour of the system under these circumstances so that it would be possible to compare it with the channel allocation schemes that were implemented at a later stage. The aim in all cases is to distribute the channels evenly so that we can ensure a minimum uniform blocking probability. P (B) A = P (B) B = P (B) C Equation [6.25] The most important ingredient in this recipe is the constant of proportionality of offered traffic per square unit area. To find this constant of proportionality, a number of simulations were performed varying different parameters and observing the blocking probabilities. The following table lists the simulations performed along with the parameters varied. A selection of these will be described and analysed in the following sections. Table 6.4 List of simulations performed No. Simulation Title Parameters Varied 1 A B C Blocking Probability vs Overlapping Radius R 2 A B C Blocking Probability vs Channels in the System CH T 3 A B C Blocking Probability vs R, CH T (Overlapping Radius & Number of Channels) 4 A B C Blocking Probability vs Offered Traffic Offered Traffic 5 A B C Blocking Probability vs R, Offered Traffic (Overlapping Radius & Offered Traffic) 6 Investigation for Uniform Blocking Probability CH A CH B 177

179 CHAPTER 6. Fixed Channel Allocation Based Techniques For both CACOS and NOAS simulations performed, the size of the normalised cells considered, has an area of 3.14 km 2 (since they are normalised with R e = 1). The number of erlangs per normalised cell was set at 200 erlang/cell. Also, the frequency reuse factor (K) considered in the system was set to 7. So, a total number of 100 channels per cell were assumed when having 700 channels in total for the whole system. These parameters have been decided upon numerical investigation, when a problem occurred during the initial simulations. The problem was that when assigning a smaller OT per cell, the number of channel per normalised cell was also considerably decreased. As a result, the distribution of the channels was not feasible since the number of channels was inadequate to be allocated based on the size of the area. To give an example, let us assume that the frequency reuse factor K is equal to 3. If a system is using 90 channels, then each cell will be allocated 30 channels. Calculating the overlapping areas, the channels allocated will depend on the size of the areas. If one area has a very small percentage of the total coverage area of a cell, then it is possible that this area might not be assigned any channel. In order to ameliorate this problem, the total number of channels and erlangs per normalised cell was set higher. Table 6.5 Simulation Parameters used Parameter Type Parameter Value Erlangs per Cell (offered traffic) 200 Frequency reuse factor (K) 7 Total number of Channels (CH T ) 700 Normalised Overlapping Radius (R/R e ) Simulation results showed that while running the simulation, some of the blocking probability plots had a saw-tooth like shape (see Figure 6.20). This is because of the way the simulation was developed and more specifically the fact that the number of channels allocated in Equation [6.22] and Equation [6.23] were floored. Results are compared with the case where we do not have cell overlap. This is the case we have a single cell with a nominal number of channels and considering an ErlangB traffic distribution. 178

180 CHAPTER 6. Fixed Channel Allocation Based Techniques Simulation Results for CACOS Scheme This simulation investigated the relationship between the blocking probability and the degree of overlap for the internal cell. As mentioned before, an internal cell consists of 6B Small, 6C and 1A area. Varying the radius of the overlapping (circular) cells, the areas A, B and C are also varied. As a result, the channels allocated vary as well. The number of erlangs per square unit was set to 200 per cell and the frequency reuse factors (K) considered for the system was 7. The radius of the overlapping cells was varied from 1 up to 1.5R e (where R e is the radius of the circle that encloses the hexagonal cell). Based on the first assumption as well as the parameters outlined above, the following plots were generated when the cluster size was set to 7. Table 6.6 Basic Simulation Parameters CACOS No. Parameter Value Units 1 Type of Cell Internal (6 neighbours) 2 K 7 3 Erlangs in Target Cell Total Number of Channels Number of Channels in a cell R e 1 unit 7 Normalised Overlapping Radius (R/R e ) unit 8 Channel Assignment Case CACOS 179

181 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.20 Individual blocking probability vs Overlap Radius (top), Total Blocking probability vs Overlap Radius (lower) Relating the individual blocking probability plot (Figure 6.20-top plot) with the variation of area A, B and C while increasing the overlapping radius (Figure 6.21), it is noticeable that as the area increases the blocking probability decreases or vice versa. Also, it is observed that the blocking probability in areas B and C is significantly lower than area A. Area A on the other hand suffers high blocking levels. Figure top plot shows that the blocking in area A is even worse than the no cell overlap case (considering just a singe cell). This applies in most times except the early stages of cell overlap. As mentioned before, the number of channels allocated in each area is based on the size of the area. So when R is varying between units the size of area C is very small. CACOS scheme floors the channels allocated in region B and C and assigns the rest of the channels in region A so that no channels are left unused. In the case of Area C when the number of channels allocated in the area is not integer e.g. 0.8 channels, the number of channels allocated is floored to zero and the blocking is 1. The total blocking probability illustrated in Figure 6.20 (lower plot) represents the general blocking all around the cell. From the results the overall blocking levels have significantly been reduced. Plot shows that the higher the cell-overlap the lower the blocking levels. Also the total 180

182 CHAPTER 6. Fixed Channel Allocation Based Techniques blocking probability is significantly lower for the CACOS scheme. The following equation describes the general blocking probability. P Blocking Total P A Area A P + Cell Area Blocking A Area B P + Cell Area Blocking Blocking A AreaC = Equation [6.26] Cell Area However, it is not indicating directly if one of the areas (A, B or C) has a high blocking probability. It is nevertheless a generalised form of representation of the blocking probability in a cell with respect to the degree of overlapping. Figure 6.21 Inner cell consists 6B s, 6C s and one area A The incremental step used for the normalised radius R show in Figure 6.20 was set to units. Increasing the resolution of the plot by further decreasing the value of the step, the sawtooth plot shape will become more obvious. This is because the floor function will be performed several times. Simulation Results for NOAS Scheme For this test the code was modified to implement the second assumption. For this case the channels allocated in area B and C were divided by 2 and 3 respectively. This means that when 181

183 CHAPTER 6. Fixed Channel Allocation Based Techniques the overlap occurs, the total number of channels in area B and C will be doubled and tripled respectively. As a result, the number of channels in area B and C of 6-neighbouring cells will be the same as if the overlapping did not take place. The difference with the previous technique is that is saving more channels for area A whilst area B and C can take advantage of the channel availability due to cell-overlap. Table 6.7 Basic Simulation Parameters NOAS No. Parameter Value Units 1 Type of Cell Internal (6 neighbours) 2 K 7 3 Erlangs in Target Cell Total Number of Channels Number of Channels in a cell R e 1 unit 7 Normalised Overlapping Radius (R/R e ) unit 8 Channel Assignment Case NOAS Figure 6.22 Individual blocking probability vs Overlap Radius (upper), Total Blocking probability vs Overlap Radius (lower) NOAS As before, the individual blocking probability as well as the overall blocking probability were monitored and compared with the case of no overlap (considering a single cell). Figure

184 CHAPTER 6. Fixed Channel Allocation Based Techniques (upper) shows that the blocking probability achieved in area A has been reduced significantly. However this was at the expense of increased blocking in both area B and C. Nevertheless, the total blocking probability is lower than the case with no overlap. Discussion of results for CACOS and NOAS Comparing CACOS and NOAS schemes, it can be clearly seen that reducing the total number of channels in the overlapping areas to improve blocking levels in the non-overlapping areas (NOAS), results in higher blocking. On the other hand simple channel allocation based on the size of the areas (CACOS) has delivered much lower blocking for the overlapping areas (B and C). The blocking probability however in area A, seemed to increase at a higher rate. This is because the size of area A decreases at a higher rate than area B (since area C is increasing). To conclude, the degree of overlap along with the total number of channels available in the three types of areas (with and without overlap) must be considered in such a way in order to ensure that the individual blocking probability will be maintained uniform (and therefore minimum) in all three areas in every cell. This criterion is more applicable in a uniformly distributed traffic environment where the calls per square unit are approximately the same Numerical Calculation for Uniform Blocking (NCUB) The calculation of the number of channels required for each area will now be performed numerically. Recalling Equation [6.17], Equation [6.18] and Equation [6.19] the blocking probabilities for areas A, B and C are a function of the offered traffic and the numbers of channels assigned. As in CACOS, channels in each cell are allocated proportional to the areas covered. Since area B and C are overlapping areas, the number of channels assigned in these areas will be doubled (in the case of area B) and tripled (in the case of area C) due to the cell overlap. Therefore the users within these two areas will be able to choose from the channels offered from any overlapping cell, which is forming this area. As a result, the blocking probability will decrease significantly. As mentioned before, the aim is to distribute the channels evenly so that the blocking probability will be the same in all areas. For the Numerical Calculation for Uniform Blocking (NCUB) scheme, area A and B were assigned different numbers of channels. Area C on the other hand was assigned the remaining channels in the cell. Base on Equation [6.20] CH A and CH B can be defined as: 183

185 CHAPTER 6. Fixed Channel Allocation Based Techniques CH A = {1 CH T } Equation [6.27] CH B = {1 CH T } Equation [6.28] and CH C = CH T - (CH A + CH B ) Equation [6.29] This investigation was repeated for different values of overlapping radius and the cell considered was an internal cell consisted of 6B Small, 6C and 1A areas. The aim was to investigate which combination of channels was giving a uniform and therefore minimum blocking probability over the whole cell. This investigation was numerical and a number of iterations were required in order to identify the optimum set of channels to be assigned in these areas. In order to implement this, the difference between the maximum and the minimum blocking of area A, B and C for a range of channels and with respect to the radius of the overlapping circles (which is set to be constant) were calculated. Using these values, it has been possible to find the minimum difference between all three areas. A table with the number of channels that had to be assigned in each area was then recorded. The following table lists the set of parameters used for the implementation of this simulation. Table 6.8 Basic Simulation Parameters Numerical Calculation of Channel Assignment No. Parameter Value Units 1 Type of Cell Internal (6 neighbours) 2 K 7 3 Erlangs in Target Cell Total Number of Channels Number of Channels in a cell R e 1 unit 7 Normalised Overlapping Radius (R/R e ) 1.3 unit 8 Channel Assignment Case NCUB This simulation was performed for the case of R = 1.3. The number of channels in area A and B were varied using a for loop so that all possible combinations were tried. Figure 6.23 depicts 184

186 CHAPTER 6. Fixed Channel Allocation Based Techniques the result of this simulation for this case. As you can see, there is a point where the difference between the values of blocking probability of every area is a minimum. Another point to note here is that a large part of the plot indicates a difference of blocking of almost 1. The reason for this was due to the negative number of channels allocated in area C when the sum of the number channels in area A and B exceeded the total number of channels in the system. Figure 6.23 Maximum Minimum Blocking probability between areas A, B and C (Difference between the maximum and the minimum blocking of area A, B and C for a range of channels) The following step was to search for the minimum point of this surface plot. This would give us the combination of channels for normalised radius equal to 1.3, while giving the minimum channels allocated to each area and this is shown in Table 6.9. Table 6.9 Optimum channel allocation for R=1.3 Areas Number of channels Area (%) A B C Total

187 CHAPTER 6. Fixed Channel Allocation Based Techniques To further investigate these results, the initial simulation was re-run but this time the number of channels per area was kept constant according to the optimum channel allocation for overlapping radius 1.3. Varying the radius of the cells, it was expected that the optimum blocking probability point, where all three areas had approximately the same blocking, would be at about 1.3. The following plot depicts the result of this simulation and verifies the statement above. Figure 6.24 Verification of the channel allocation technique. Optimum overlap radius is equal to 1.3 From Figure 6.24, it can be clearly seen that the blocking probability is the same for all three cases, when normalised radius is approximately 1.3. This value is considered to be optimum since the channel allocation at this point ensures minimum uniform blocking along areas A, B and C. From the plot, we can see that for the same channel allocation and for a range from , the results satisfy the previous statement. This means the same combination of channel allocation could be used for a small range of radiuses. 186

188 CHAPTER 6. Fixed Channel Allocation Based Techniques The practical significance of the results presented above, lie on the fact that a HAP communication system might require to reduce the size of its cells during its service hours to extend its batteries lifetime. To give an example a HAP might have to reduce its transmit power on a particularly windy day that had to use a lot of its power to do station keeping. In order to define the limits of the overlapping radius range, for which the same number of channels could guarantee uniform and therefore minimum blocking in these areas, a new simulation technique was devised. In this simulation, the optimum channel allocation was defined while varying the radius. The rate of increase in the radius was varied, in order to observe the channel allocation changes in sufficient detail. Running the simulation incrementing the overlapping distance using various steps proved that the channel allocation changed approximately every (R=) So, the step used was set to The other parameters used in the simulation are shown in Table Table 6.10 Basic Simulation Parameters for the Optimisation Technique No. Parameter Value Units 1 Type of Cell Internal (6 neighbours) 2 K 7 3 Erlangs in Target Cell Total Number of Channels Number of Channels in a cell R e 1 unit 7 Normalised Overlapping Radius (R/R e ) unit 8 Channel Assignment Case NCUB 9 Type of Cell Internal (6 neighbours) Figure 6.25 shows the number of channels required in a cell (internal cell case see Figure 6.10) for a specific size of radius in order to guarantee a minimum and uniform blocking all over the cell. 187

189 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.25 Channel Allocation based on overlapping radius Figure 6.26 illustrates the difference between the maximum and minimum blocking probabilities of areas A, B and C while varying the normalised overlapping cell radius from It can be clearly seen that when there is a large amount of cell overlap there is sufficient flexibility to minimise the blocking probability difference between areas. It is nevertheless known that the size of area C starts to get formed just after R becomes greater than 1. At this stage, the size of area C is very small (about 2%) and so blocking probability is sensitive to any increase or decrease of the channels leading to greater differences in blocking probability between the areas. Figure 6.26 MAX MIN blocking function versus normalised overlap radius. Normalised to R e 188

190 CHAPTER 6. Fixed Channel Allocation Based Techniques To bridge the large differences of blocking in the areas the cell overlap can either be increased or the number of channels per cell can be increased. The second case might be impractical and expensive as we are trying to solve the problem by just bringing in more channels. The first case however, is more interesting since it does not require buying more spectrum to improve the system capacity. A plot comparing the cell overlap case with the non-overlapping case is shown in Figure This plot depicts the average blocking probability of a non-overlapping cell (isolated) and an overlapping cell. The non-overlapping cell is considered to be one area, so the number of channels allocated in this area (cell) is constant, and so the blocking probability does not change with normalised radius. In the overlapping case, the number of channels allocated to area B and C varies in proportion to the area and there is twice and three times the number of channels available in areas B and C respectively. This extra freedom means that the overall blocking probability is much lower in this case. Figure 6.27 Average blocking probability in an overlap cell and in a non-overlap Cell. Normalised to R e NCUB optimisation can be performed continuously to maintain an acceptable level for the QoS. This can be done at a user level or on the HAP provided that there is enough power to cope with all the signalling and power processing. Further investigation and performance analysis of the NCUB optimisation technique At the beginning of this section, it was stated that the optimisation performed was based on a constant radius and constant offered traffic. The aim now is to explore the ability of the system to cope with changing conditions such as radius and traffic while being optimised for fixed conditions. It is important to explore the levels of tolerance of these variations in order to ensure 189

191 CHAPTER 6. Fixed Channel Allocation Based Techniques uniform QoS all over the coverage area and to stress the capabilities of the system in order to find out how well it can perform. To do so, various conditions had to be altered. All the scenarios are described in detail. Table 6.11 lists the scenarios that have been investigated. Table 6.11 Optimisation Technique Performance evaluation tests Test No Description of test The number of channels and the offered traffic increased while keeping the ratio between them constant. The size of the overlap radius varied while keeping the optimised set of channels constant. The erlang per cell increased while keeping the optimised set of channels constant. The erlang per cell increased while monitoring the channel distribution for different radii. The main characteristic of this model is that the cell of interest (centre cell) is always examined in isolation. Also when referring to the number of erlangs in the target cell we mean the cell under examination which in this case is the centre cell (with 6 neighbours). The neighbouring cells however are assumed to have no traffic in order to be able to provide the channels required in the overlap regions of the target cell, should they be required. Also, when referring to channels per cell in the simulation, it means the available channels in the target cell. However, when taking into account the channels in the overlap regions borrowed from the neighbouring cells, the total number of channels available in the target cell will be much greater than what has been initially defined as channels per cell. The amount of increase of the channels is defined from the degree of overlap. If for example the overlap radius R is 1.5, then area C will be at its maximum size (see Figure 6.21). Therefore, the total number of channels allocated in this area from the target cells is tripled due to the overlap. As a result, the total number of channels in the cell will be at its maximum and therefore the overall blocking will decrease significantly. This has been a very basic model used to implement various generic techniques in order to explore cell overlap. As will be seen in the following chapter, this model has been very useful when trying to verify the Monte Carlo based simulation model. The Monte Carlo type model performs the same task but in a more detailed and realistic way allowing more flexibility to the investigation of efficient channel allocation (more details on the Monte Carlo Model are given in the following chapter). 190

192 CHAPTER 6. Fixed Channel Allocation Based Techniques To summarise the assumptions made: Traffic is generated only in the target cell. The offered traffic per square unit is constant. The total number of channels in the target cell varies according to the level of overlap. To summarise the conclusions made: This model was used to illustrate the benefits of cell overlap rather than being used for any practical implementation. It is implemented to explore cell overlap and provide a way to verify the correct operation of the Monte Carlo based simulation model. Test 1: For this test the number of channels in the system has been increased along with the offered traffic. This is to show how much better the optimisation technique can perform with these conditions since the channel distribution will be more accurate; bearing in mind that the number of channels assigned in an area has to be an integer. This is not a condition that will be varying during the system s operational lifetime but it is important to set the right number of channels that will generate acceptable levels of blocking for other varying conditions (such as OT). The simulation is performed twice. In the first run the number of Channels per Cell / Erlangs in Target Cell is equal to 200 / 300 and in the second one 300 / 450 maintaining a ratio of 2:3. The aim in all cases is to maintain as low as possible uniform blocking all over the coverage area. Figure 6.28 illustrates an example of discrepancy between the three regions forming a cell. Discrepancy of the blocking probability between area A, B and C Figure 6.28 Blocking Probability Discrepancy between areas The results in Figure 6.29 show that the blocking probability in each area is now closer to each other while the overlap radius varies. Still, the difference between the blocking of the areas within the same cell is significant. This is because there is small number of channels allocated per cell. Since the distribution of channels performed assumes an integer number of channels, it 191

193 CHAPTER 6. Fixed Channel Allocation Based Techniques is more difficult to accurately divide and allocate the channels to the areas. The levels of blocking are also different between the four types of cells. This was expected since the size of the overlap areas is different in each case. Table 6.12 NCUB optimisation Technique Case of 200 Channels / Cell Type of Cell 6 neighbours Erlangs in Target Cell 300 Number of Channels 200 R e 1 Overlapping Radius (R) Figure 6.29 Area Blocking and Cell Blocking after optimisation of 200 channels per cell and OT of 300 per cell Given that an integer number of channels must be maintained and that a more accurate distribution must be achieved, the number of channels per target cell along with the offered traffic has been increased while the ratio between these two was kept constant. By doing this, we would expect that the trunking efficiency will improve and the discrepancy between the blocking of the areas will become smaller. The following set of plots was generated for

194 CHAPTER 6. Fixed Channel Allocation Based Techniques channels per cell and an OT of 450 erlang per cell. The ratio between the OT and the number of channels per cell was kept constant for comparison with the previous simulation. Table 6.13 Optimisation Technique Case of 300 Channels / Cell Type of Cell 6 neighbours Erlangs in Target Cell 450 Number of Channels / Normalised Cell 300 R e 1 Overlapping Radius (R) Figure 6.30 Area Blocking and Cell Blocking after optimisation of 300 channels per cell and OT of 450 per cell From Figure 6.30 it can be deduced that the discrepancy between the blocking of the area has decreased. The reason for this is because we now have more channels to distribute into the various areas and although the ratio between the channels per cell and erlang per cell remains the same, the division performed is more accurate (remember that the number of channels allocated has to be an integer). In the following section the sensitivity of the system to the number of channels available in an area is investigated. This is to quantify the sensitivity of the NCUB optimisation technique to the total number of channels available as well as to stress the significance of the trunking efficiency to a communication system. 193

195 CHAPTER 6. Fixed Channel Allocation Based Techniques 6.6. Sensitivity to numbers of channels per area As seen in this Chapter, the concept of blocking probability has been used to quantify the performance of a connection oriented multi-user system. The blocking probability has been calculated using the ErlangB formula shown in Equation [6.7]. Through this formula, it is possible to tell the traffic load that can be supported by the system when a blocking probability requirement is set and a specific number of channels is available. Trunking Efficiency in [68] has been expressed as the value of the traffic load per channel in units or Erlangs/channel. It has also been shown that a system with greater number of channels is able to support a greater traffic load per channel. To illustrate the relation between the number of channels and number of erlangs per cell, the following plot of ErlangB blocking has been generated. The ratio between the channels and erlangs per cell was set to be equal. Figure 6.31 Trunking efficiency plot based on erlangb formula. The number of Channels per Cell in this case is equal to the number of Erlangs per Cell Figure 6.31 illustrates that by increasing the number of channels and erlangs per cell while keeping the ratio between them constant, the blocking probability decreases significantly. As a result, the variation of the blocking probability becomes very small when increasing these two factors. This means that the trunking efficiency improves as the number of channels increases. This comes to support the above statement [68] saying that the system with greater number of 194

196 CHAPTER 6. Fixed Channel Allocation Based Techniques channels can support a greater traffic load per channel. Although this points out that it is better having an ever increasing number of smaller channels (since the spectrum available is limited), it contradicts with the Shannon Equation [69] which shows that a narrow-bandwidth channel is unlikely to support high bit-rate modulation schemes. From Figure 6.31, it can also be stated that for a small number of channels, the blocking is highly dependent on the exact number of the channels. So if for example the number of channels is incremented by one, the blocking probability in the area will decrease significantly. This shows shows the significance of trunking efficiency in the channel optimisation technique presented in Section 6.5. To investigate the system behaviour in more depth, a plot has been drawn which illustrates the degree of variation when the number of channels allocated in an area, is increased or decreased by one channel. For a small number of channels, the quantisation error is extremely high leading us to the conclusion that the optimisation technique would not be able to match the right number of channels for every region. To give an example, a channel is being taken from one region and is being assigned to another when there are 10 channels in the cell. As a result, the channel variation in each region will be 10%. This will have a direct impact at the blocking levels of the regions. % variation of +/- 1 Channel Number of Channels Figure 6.32 Quantisation Error Plot defines the margins the optimisation technique can operate 195

197 CHAPTER 6. Fixed Channel Allocation Based Techniques What can be deduced from this test is that the optimisation technique performs better when having more channels per cell as the trunking efficiency is much higher for larger values of number of channels. Based on the previous test, we now investigate the significance of having a large number of channels in an area. In this test, the blocking probability is maintained constant while the number of channels and the offered traffic are varied. The aim here is to show the level of change in the blocking probability when increasing and decreasing the number of channels in the cell by one (always for different number of channels and levels of OT). As been stated before, the number of channels allocated in an area must be an integer. Thus the number of channels in a cell can be increased or decreased by a minimum of one channel. In this model, the number of channels increases along with the OT in order to maintain a constant blocking which is represented as the centre (continuous) line in Figure 6.33 and Figure While performing this, a channel has been added or subtracted from the total number of channels calculated to maintain the desired blocking probability. The following plot depicts the level of change in the blocking while adding or subtracting a channel from the total number of channels. Blocking Probability -1 Channel +1 Channel Number of Channels Figure 6.33 (+1 / -1) Channel variation and the effect on the desired blocking probability (which is 0.05 in this example) From Figure 6.33 it can be deduced that the level of change in the blocking is greater when subtracting a channel rather than adding. It can also be seen that if not impossible it is extremely 196

198 CHAPTER 6. Fixed Channel Allocation Based Techniques difficult for the optimisation technique to accurately allocate the channels into the areas when the total number of channels is very small. If for example there are 40 channels in the area and the blocking probability is maintained at 0.05 (5%), then by adding or subtracting a channel from the total number of channels it can cause a change in the blocking probability of about +/- 20%. To illustrate the degree of change in the blocking probability in a more generalised form, the previous plot was based on percentages. % increase in Blocking Probability -1 Decrease +1 Increase Number of Channels Figure 6.34 (+1 / -1) Channel variation and the effect on the blocking probability as a percentage of the desired blocking (which is 0.05 in this example) As mentioned before, the aim of the optimisation technique is to maintain minimum while uniform blocking between the three areas A, B and C. However, in order to perform this task as accurate as possible, it is essential that the total number of channels in the regions must be large enough. As a result, when moving a channel from one region to another the blocking will not change in great steps but in small steps so that a minimum discrepancy can be achieved. Test 2: The next parameter altered was the radius of the cells and hence the degree of overlap. The aim here was to explore the ability of the system to cope with the changing radius while the system was optimised for a fixed one. Although is practically difficult but not impossible to dynamically vary the cell radius while the HAP is in service, it is useful to know what the 197

199 CHAPTER 6. Fixed Channel Allocation Based Techniques results will be. In this example the overlapping radius R is normalised based on R e (see Figure 6.4). Table 6.14 Optimisation Technique Case of the varying radius No. Parameter Value Units 1 Type of Cell 6 neighbours 2 Erlangs in Target Cell Number of Channels / Normalised Cell R e 1 5 Normalised Overlapping Radius (R/R e ) (steps of 0.005) units 6 Optimised Overlapping Radius (R) 1.25 The channel optimisation this time was performed when the radius was set to be at 1.25 units. Table 6.15 lists the channels required for this radius. Table 6.15 Optimum channel allocation for R=1.25 Areas Number of channels Area (%) A B C Total

200 CHAPTER 6. Fixed Channel Allocation Based Techniques Figure 6.35 Area Blocking and Cell Blocking optimised at R=1.25 for 200 channels per cell The optimisation in this example has been performed for the case where radius R is 1.25 units. The results presented in Figure 6.35 show that the minimum while uniform blocking occurs only when the overlap radius is 1.25 units. Regardless of whether the radius increases or decreases from its initial optimised size, the blocking uniformity is lost. It can therefore be stated that changing the radius of the cells requires re-optimisation of the channel allocation especially when considering dynamic changing of the overlap radius. Changing the radii of the cells, also changes the sizes of the regions. As a result, the channel allocation is not optimum anymore. If for example area C is being allocated 10 channels when R is 1.1 units, then by keeping the same number of channels when R is 1.5 units it means high blocking since the size of the region increased. This test has verified that the optimisation technique performs well when performed and that is necessary at any instant the overlap radius has changed. If it is not performed, then the blocking in the regions will be uneven and the system will not perform in an optimised manner. Test 3: Varying the offered traffic is the other condition that could possibly cause inefficient operation of the optimisation technique leading to re-optimisation. The question that arises from this is which degree the offered traffic per cell has to be changed in order to perform reoptimisation of the channel allocation. In order to investigate this, the optimisation was 199

201 CHAPTER 6. Fixed Channel Allocation Based Techniques performed again at R = 1.25 units for 200 channels per cell and 300 erlang per cell. This time, the OT per cell varied in small steps from 200 to 400 erlang per cell. Blocking Discrepancy : 2% Figure 6.36 Area Blocking after optimisation base on 200 channels per cell and OT of 300 per cell Figure 6.36 depicts the individual area based blocking as well as the average blocking in the system when varying the OT. From this figure, it can be clearly seen that the blocking increases as well as the discrepancy when increasing the OT. Furthermore, the overall blocking is always smaller than when no cell overlap is permitted. The different sizes of the areas cause problems to the optimisation technique when trying to allocate them channels at higher OT. The larger the area is the higher the OT it will experience especially when increasing the OT. Since the number of channels is fixed (optimised according to Table 6.15) for all three areas, the blocking is bound to increase more in the areas with bigger area. Figure 6.35 and Figure 6.36 show that it is important to define a set of limits where the reoptimisation of the channels will be required. In the case of the varying cell radius, the discrepancy between the blocking of the areas increases significantly and therefore the limits will be very close. This means that re-optimisation of the channel distribution will be required every time the size of the cell changes. In the case of the varying OT, it is not necessary to optimise the channel distribution very often. When the OT is decreased, the blocking in all regions is dropped significantly so that any discrepancy can be tolerated. In the case where the 200

202 CHAPTER 6. Fixed Channel Allocation Based Techniques OT increased, the discrepancy became more obvious and therefore re-optimisation would be necessary. The limit to perform re-optimisation in this case is when the discrepancy exceeds a certain threshold that the service provider has set (see Figure 6.36). So if for example the maximum possible discrepancy allowed is when the blocking becomes equal or greater to 2%, then re-optimisation will be required. In the example illustrated in Figure 6.36, this will happen when the OT reaches 355 erlangs. Performing optimisation ensures uniform blocking while the overall blocking will increase (assuming that only traffic increases while the number of channels is fixed). Test 4: A simulation model has been developed in order to examine the percentages of the channels being allocated in the areas when performing the optimisation technique, every time the OT changes. The number of channels per target cell was kept constant (300 channels per target cell) while the OT varied from erlangs per cells. For every value of OT, the channel allocation was re-optimised. The following plots were generated for overlap radius R = in steps of 0.05 for the case of 6 neighbouring cells. In each case the overlap radius is kept constant while the OT varies. It is useful at this point to recall the plot that describes the size of the areas when the overlap varies for the case of the 6 neighbouring cells depicted in Figure 6.21 Relating the plot in Figure 6.21 and the plots in Figure 6.37, it can be seen that the allocation of channels in each area increases with respect to the size of the area. For example, area C is allocated a significant number of channels when R>1.2. As mentioned above, while the OT varies the size of all the areas remain constant. As a result, the OT varies in proportion and therefore the percentage of the OT allocated per area is also constant with respect to the total OT per cell. (The OT assigned per area is proportional to the size of the area. Since the overlap remains constant, the OT will also be in proportion to the size of the areas). The allocation of the channels however, changes as the OT increases. For area C, the number of channels assigned increases along with the increasing OT, whereas for area A it is constantly decreasing. Also the number of channels allocated in area B is directly proportional to OT except when R = 1.4 and

203 CHAPTER 6. Fixed Channel Allocation Based Techniques R = 1.25 R = 1.35 R = 1.4 R = 1.5 Figure 6.37 Area channel allocation (%) vs OT 6.7. Summary and Conclusions In this Chapter the FCA scheme has been briefly presented followed by a detailed mathematical analysis of the cellular overlap experienced by a HAP. Based on this investigation, various fixed channel allocation based schemes have been developed and tested. These schemes were based on a theoretical model for calculating blocking probability, which use the ErlangB formula to effectively represent average values of blocking. The core simulation model was based on an erlangb distribution; i.e. an infinite number of users and an average offered traffic per square unit area. The disadvantage of this model is that it is not flexible to implement complex channel allocation schemes based on a finite number of users because it was designed to be simple and generate basic but fundamental results. These results have been important to prove the 202

204 CHAPTER 6. Fixed Channel Allocation Based Techniques significance of cell overlap and justify further investigation. Although these generic techniques cannot be directly used for a HAP system, we will use the understanding gained through the simulation results as a basis for practical schemes which will be implemented in chapter 7. Therefore, these schemes contain the basis to explore cell overlap. From the results presented in this chapter, it is shown that exploiting cell overlap improves the QoS in terms of the overall blocking probability. It has also been shown that because of the overlap the blocking across the different types of areas of overlap formed is uneven and therefore not fair. An optimisation technique can be used to subdivide the channels into the areas. NCUB scheme manage to bridge the gap of blocking difference between the areas and thus provide fair service (uniform blocking). Nevertheless its performance depends on the number of channels available per cell. More specifically, it has been shown that it is extremely difficult for the NCUB optimisation technique to accurately allocate the channels into the areas when the total number of channels was very small. In Figure 6.33 it has be shown that if for example there are 40 channels in the area and the blocking probability is maintained at 0.05 (5%), then by adding or subtracting a channel from the total number of channel it can cause a change in the blocking probability of about +/- 20%. Re-optimisation is required on a regular basis and this might happen when the radius has changed (see Figure 6.35) e.g. reduced to conserve energy or when the OT has increased (Figure 6.36). The interval that the optimisation must occur depends on the discrepancy of the blocking levels between the areas. 203

205 CHAPTER 7. Providing fair services whilst exploiting cell overlap Chapter 7. Providing fair services whilst exploiting cell overlap 7.1 Introduction Channel Allocation Schemes Area Based Fixed Channel Allocation Scheme (ABFCA) Region Based Fixed Channel Allocation Scheme (RBFCA) Uniform Fixed Channel Allocation Scheme (UFCA) Uniform Fixed Channel Allocation - II (UFCA - II) Summary and Conclusions Introduction The purpose of this Chapter is to look at a number of Fixed Channel Allocation (FCA) schemes, the benefits of cell overlap and how it can be ensured fair access to the actual resources. In chapter 6 the theoretical split of channels that should be allocated to the different areas based on the ErlangB formula has been examined. In this chapter a set of more practical-based channel allocation schemes are examined to investigate ways to improve system performance while maintain a fair service across the coverage area. In Section 7.2 the basic FCA scheme is presented. Then in Section 7.3 the Area Based Fixed Channel allocation scheme (ABFCA) employing cell overlap is presented. ABFCA proves that by exploiting cell overlap, the system performance improves but results show that the Quality of Service (QoS) is not uniform across the cells. Then in Section 7.4 the Regional Based Fixed Channel Allocation Scheme RBFCA is presented. RBFCA has been implemented in order to unify the blocking probability levels between the regions. The performance of RBFCA was much worse than the case of FCA scheme with no overlap. Following, the Uniform Fixed Channel Allocation (UFCA) model is presented which is based on the Random Acceptance Factor (RAF) and it indented to achieve both uniform blocking across the cell as well as it achieves better performance than the simple FCA scheme which does not employ cell overlap. Following, the UFCA-II is presented. UFCA-II is based on UFCA scheme. Nevertheless, apart from ensuring uniform blocking level, it also ensures uniform data rates across the area. Finally the conclusions are presented. 204

206 CHAPTER 7. Providing fair services whilst exploiting cell overlap 7.2. Channel Allocation Schemes To ensure realistic results, 37 circular cells were used to eliminate the problem of edge-effect [70]. Statistics were only collected from the centre cell and this is because the majority of the cells in a 121-cell scenario have six neighbours (e.g. centre cell). Each of these 37 cells uses a different group of 30 channels within its coverage area and therefore co-channel interference is not present. We have also assumed that there is direct line-of-sight communication between the user and the HAP. 100,000 users have been assumed with 500,000 conversations to be made. The arrival process is a Poisson distribution and the length of the phone calls has a negative exponential distribution. The users are uniformly distributed within the coverage area, and the offered traffic (OT) is quantified in terms of Erlangs per square unit area. The total OT in the 37-cell coverage area ranges from 8-10 Erlang per square unit times the actual size of the coverage area (one unit of length is taken to be the external radius of the cells, R e ). Based on Region B (No Area C) found in equation C.2 in Appendix C and Region C found in equation C.16 in Appendix C, the size of the total size of coverage area (CA) for 37 circular cells is equal to: ( 90 Region B Region C) 2 CA = 37π R (No Area C) 54 Equation [7.1] Assuming that R e is equal to 1 unit, then R i is equal to 3 units, and the total coverage area is 2 equal to 99.9 square units. This value remains fixed for any overlap radius despite the fact the cell radius might change. This is to ensure constant OT within the coverage area. The same parameters have been used in all channel allocation models with the only exception the way the channels are allocated. To summarise, Table 7.1 Simulation Parameters for FCA schemes No. Parameter Value Units 1 Number of Cells (c) 37 2 Channels per Cell 30 3 Offered Traffic 8-10 erlang/sq. unit area 4 Users 100,000 5 Conversations to end of simulation 500,000 6 Coverage Area 99.9 sq.units 7 Average Call Length 6 Minutes 8 Arrival Process Poisson 205

207 CHAPTER 7. Providing fair services whilst exploiting cell overlap Fixed Channel Allocation with no Cell Overlap To comment on the performance of the newly implemented channel allocation schemes based on overlap we first needed to compare them with a fixed channel allocation scheme that does not exploit cell overlap. Based on the parameters mentioned above a basic channel allocation scheme was developed. Description The Fixed Channel Allocation (FCA) scheme was implemented such that each cell was allocated a fixed set of channels and only allowed users to be allocated a channel from the nearest base station. No overlap was considered and hence no choices were made by the users about which cell should allocate them a channel. Performance Figure 7.1 depicts how the standard FCA scheme performs against offered traffic (OT). It clearly shows that the system can maintain a 4% blocking probability for an OT less than 9.5 Erlang per square unit. As mentioned before, the users in this scheme could only be connected to the closest base station. No overlap Figure 7.1 Nearest Base Station Scheme - No Overlap Considered 206

208 CHAPTER 7. Providing fair services whilst exploiting cell overlap 7.3. Area Based Fixed Channel Allocation Scheme (ABFCA) Description The ABFCA scheme is based on the standard FCA scheme presented above. As before, each cell has a fixed number of channels, and can allocate them to any user within its coverage area. However, the cell radius (R) is now equal to 1.25 times the initial cell radius (R e ). This means that users positioned within radius R of the centre of any cell can connect to this cell. The users will first search for the number of cells they can connect to (up to 3 cells) and then they will pick a channel from whichever cell has the most available channels. If, for example, a user happens to be in area C, he/she will pick a channel from the one out of three cells within range that has most channels available. Performance In this simulation the radius of the cells defining the degree of overlap is set to R=1.25. Notice that although the overlap radius has been increased, the OT within the total coverage area remains the same (as in the previous case with no overlap). Figure 7.2 depicts the blocking levels for areas A, B and C in the centre cell for the case where we consider cell overlap, in contrast to the standard FCA scheme. From the plot, it can be seen that overall (considering the total cell blocking levels), the overlap scheme performs better than the one with no overlap (previous scheme); however the blocking levels for the users positioned in area A, B and C vary considerably. Users in area A experience worse blocking levels than the users in area B and C. For areas B and C, where the users have the option to choose from more than one cell, the blocking levels are much lower than the case with no overlap. 207

209 CHAPTER 7. Providing fair services whilst exploiting cell overlap No overlap Area A Area B Area C Figure 7.2 Area Based FCA Scheme (ABFCA). From the results above we can verify that the areas formed by the overlap of two or more cells enjoy the advantage of higher trunking efficiency, and as a result, the blocking levels are much lower than the case where no overlap is considered. Therefore, the overall blocking levels of the ABFCA scheme are much lower than the no overlap case. Nevertheless, it is not permissible to favour any area (in this case B and C) for the sake of maintaining the lowest possible overall blocking levels in the system. The ABFCA scheme however, indicates that utilising cell overlap can significantly reduce the blocking levels. In addition, there is great potential to improve the performance of ABFCA by careful analysis of the channel distribution based on the performance of the individual areas. More specifically, in this scheme it is useful to reduce the blocking levels in area A, by increasing slightly the blocking levels in area B and C. The aim is to try to do this by keeping the total cell blocking level below the non-overlap case while increasing the fairness of the scheme. The following scheme to be presented has been designed and implemented based on these criteria. 208

210 CHAPTER 7. Providing fair services whilst exploiting cell overlap 7.4. Region Based Fixed Channel Allocation Scheme (RBFCA) Description Here, we are trying to control the channel allocation on a region-by-region basis. This is implemented based on the 1 st case of channel allocation presented in Section 6.4 page 170. The total coverage area was partitioned into a number of small regions and instead of assigning a fixed number of channels within each cell we now have a size-based optimisation mechanism. This mechanism performs a numerical analysis same to NCUB described in Chapter 6 Section 6.5. The main concept when implementing this scheme was to ensure that there would be fairness in the service, in terms of uniform blocking probability. This model has been based on the ABFCA scheme with the only difference that the coverage area was partitioned into a number of small regions and instead of assigning a fixed number of channels within each cell, we had a sizebased optimisation mechanism to assign the channels into the small regions. This mechanism was developed in order to find the right combination of channels to give the smallest blocking probability while at the same time ensuring uniform blocking probability all over the cell. This required a number of iterations during which the blocking as well as the discrepancy between the areas were monitored in order to identify the optimum number of channels to be assigned in these regions. The number of channels allocated was a function of the size of the individual regions formed due to the overlap and the number of cells forming these regions. Channels assigned to a region can only be used within this region, and inter-cell channel borrowing (i.e. channels being used to another part of the cell apart from the one it is designated) is not permitted. This scheme is unable to cope when cells were allocated with a small number of channels, since having a small number of channels overall, a very small number will be allocated to each region (look Figure 6.33 Chapter 6 for degree of change). As a result, there might not be enough channels to be allocated to all regions, or some of the regions might have more channels than others, causing large non-uniformities in terms of blocking probability within the coverage area. This goes back to the fact that the channels allocated to the regions must be integers. As mentioned before, it is not possible to allocate the exact number of channels based on the size of the regions. It is therefore necessary to use the floor-function. (More on floor function can be found in Chapter 6 Section 6.6.) 209

211 CHAPTER 7. Providing fair services whilst exploiting cell overlap Area A: Region 0-6 Area B Small: Region Area B Large: Region Area C: Region Figure 7.3 Partitioning areas into regions Partitioning areas into regions has been essential in order to examine the blocking in each region rather than in each area. In the previous chapter, the channel allocation has been performed based on the total size of the areas A, B and C. However, area B and C in a cell consists of a number of regions depending on the number of neighbours the cell of interest has. This information has been useful to decide on what sort of improvements had to be made in order to develop a well performing channel allocation scheme. Performance In this simulation, some of the basic parameters used in Table 7.1 were changed. The total number of cells was set to be 7, the number of channels per cell was increased to 50 channels, the total OT in the coverage area was from 6-12 Erlang per square unit area times the actual size of the coverage area which was equal to units (for 7 cells). The radius of the cells defining the degree of overlap was set again to R=1.25. The following table lists the number of channels allocated per region based on the optimisation technique. These channels refer to the centre cell illustrated in Figure 7.3 as this is the cell where the statistics are gathered. The rest of the areas in the surrounding cells are not of any interest. 210

212 CHAPTER 7. Providing fair services whilst exploiting cell overlap Table 7.2 Channel Allocated per region in centre cell using RBFCA Region Number Region Type A B Small C Channels per Region The reason for reducing the total number of cells to 7 and not keeping it to 37 is because primarily the edge-effect problem was not an issue in this scenario and because a significant amount of time had to be spent on writing the code which would perform the channel allocation on a regional basis for 37 cells without actually producing anything different from the 7-cell case. This is based on the fact that statistics were gathered from the centre cell, and that the channels were allocated on a regional basis without being exchanged between the regions. Thus, the edge-effect problem was not an issue. Using the 7-cell scenario in this case would give the same results as the 37-cell scenario but with much less time spent in writing, running and debugging the code. Figure 7.4 depicts how the levels of blocking into the regions have become fairer than in the case of ABFCA scheme. However, the blocking levels in all regions have also increased beyond the standard FCA scheme s levels (that has been simulated again based on the new parameters). This comes as a result of partitioning the coverage area into smaller regions. Each cell now consists of different size regions and therefore the group of channels assigned to this cell is also partitioned into different smaller groups. As a result the trunking efficiency is significantly reduced. 211

213 CHAPTER 7. Providing fair services whilst exploiting cell overlap Figure 7.4 Region Based FCA Scheme (RBFCA). Overlap has improved system performance Furthermore it can be seen in Figure 7.4 that the blocking levels are not completely uniform across the cell especially at higher OT levels. This is because it is difficult to accurately divide the channels based on the offered traffic of every region Uniform Fixed Channel Allocation Scheme (UFCA) RBFCA has shown that it is possible to introduce fairness to the system but this comes at the expense of low trunking efficiency and therefore higher blocking levels. ABFCA has shown the potential in terms of its trunking efficiency and the reduced blocking probability across the cell area at the expense of non-uniform blocking across the cell area. UFCA is indented for the first time to exploit both the fairness given by the RBFCA and also the improved trunking efficiency achieved by ABFCA. Description UFCA is based on the 3 rd case of channel allocation presented in Section 6.4 page 170. Here, each cell has a fixed number of channels, which can be allocated to any user within its coverage area. In this scheme, certain restrictions are imposed in order to prevent a proportion of the channels from being allocated to the overlap areas, to allow them to remain available to areas with no overlap (area A). The parameters chosen for this scheme take into account the performance of the ABFCA scheme shown in Figure 7.2. The aim of this scheme is to improve 212

214 CHAPTER 7. Providing fair services whilst exploiting cell overlap the results illustrated in Figure 7.2, so that the blocking in all regions becomes lower than the standard FCA model. From Figure 7.2 it is apparent that more channels have to be allocated for the users in area A, and less channels for the users in areas B and C. One way of doing this without directly shifting channels from one area to the other, which requires partitioning the cell into small regions (e.g. RBFCA in Figure 7.4), is by blocking a proportion of users in area B and C even though there are channels available. The channels saved from area B and C can then be used in area A. As a result, the blocking of area A will be decreased and the blocking in area B and C will increase. The model proposed saves the last channel in area B and the last two channels in area C from being automatically used: instead these channels are only used on a random basis: when a random number generated is greater than a random acceptance factor (RAF). The RAF thus comes into effect every time a user in either area B or C requests a channel and there is only one (or two in the case of area C) left available for that area. After experimentation, it was found that this combination is proven to perform better than the other combinations tried since it was impossible to ensure uniform blocking in each region. Figure 7.5 illustrates diagrammatically how the UFCA scheme operates. Figure 7.5 UFCA RAF defines probability of acceptance in area B and C The optimum value for RAF was found, after a numerical investigation, to be a function of the level of offered traffic per unit area, and is given by: 213

215 CHAPTER 7. Providing fair services whilst exploiting cell overlap OT optimum RAF = α + β ln Equation [7.2] OTvarying where α defines the probability that the last channel will be saved by blocking the user in the area of reference (area B or C) and β is a scaling factor for the RAF to optimise it for a range of offered traffic. OT Optimum is the chosen value of offered traffic we wish to optimise the system for. Initially, β is set to zero in order to find the best probability factor α by requiring the blocking levels in all three areas to be as close as possible to each other. For example, in Figure 7.2, it can be clearly seen that the blocking of area A increases approximately exponentially. We therefore need to increase individually the blocking in area B and area C using different RAF scaling factors and as a result decrease the blocking in area A. This technique enables us to control the number of channels being allocated into certain areas without partitioning the group of channels of a cell into smaller groups, and retaining the maximum number of channels to be available for localised hot-spots in traffic demand. Performance As before, the radius of the cells defining the degree of overlap is set to R=1.25. parameters α, β and the OT Optimum used in this simulation are given in Table 7.3. The Table 7.3 α, β and OT Optimum parameters used for overlap radius R=1.25 Area Type α β B C OT Optimum 8.8 From the results shown in Figure 7.6 we can clearly see that the total cell blocking has been reduced. Furthermore, the blocking levels in all regions are approximately the same. 214

216 CHAPTER 7. Providing fair services whilst exploiting cell overlap No overlap drop in P(Block) 10.2% increase in OT Figure 7.6 Uniform FCA Scheme (UFCA). The blocking probability has been reduced In comparison with the other schemes, this model has improved the QoS by reducing the blocking probability in all areas, and has also achieved uniform blocking levels across the cell. More specifically, the blocking probability has dropped by approximately 2.5% at OT Optimum and the OT supported increased by 10.2% than in the standard case with no overlap. This improvement does not require any prior knowledge of the interference environment. The significance of the RAF is that the control mechanism of channel allocation has now changed from a limited number of channels that was before (e.g. RBFCA) to potentially unlimited granularity in terms of time. If for example there are 100 channels per cell there could be a granularity of +/-1 channel (e.g. RBFCA). With the RAF assuming an unlimited time we have an unlimited granularity and although we are restricting the number of users coming into the system, we have maintained maximum trunking efficiency. RAF gets round the assignment of discrete integer number of channels to each area or region because RAF is applied over a very large number of conversations. RAF performs the optimisation based on time which is potentially infinite rather than a restricted number of channels. It therefore controls the QoS over time rather than actually restricting channels indefinitely. Up to now, we have assumed a cell radius of R=1.25. Figure 7.7 however shows that this technique is also applicable for other overlap radii. The simulation was performed for a fixed OT of 8.8 Erlangs per square unit area and a range or radius R ranging from

217 CHAPTER 7. Providing fair services whilst exploiting cell overlap Area A Area B Area C Figure 7.7 Uniform FCA Scheme (UFCA) Blocking Probability in Centre Cell for Different Degrees of Overlap Radius. As the values of α and β for areas B and C are optimised for R=1.25, the optimisation clearly does not work for lower overlap radius as the blocking probabilities in area A, B and C are significantly different; showing that the values of α and β are functions of the amount of overlap. Although the blocking probability decreases as the cell overlap increases, a radius of greater than R=1.25 assumes that users can cope with the increased level of interference Uniform Fixed Channel Allocation - II (UFCA - II) Description The previous section in conjunction with [62], [59] and [43] have illustrated how channel assignment strategies could be configured for a High Altitude Platform architecture exploiting cell overlap. From the results it has been shown that cell overlap can be exploited to improve the performance for a scheme that uses fixed channel allocation. The areas that are served by more than one cell benefit from the higher trunking efficiency and as a result they will have much lower blocking than the areas served by one cell. In order to improve uniformity and reduce the blocking probability within a cell, a technique was developed called Uniform Fixed Channel Allocation (UFCA). As shown before, this technique has made it possible to control the number of channels being allocated to certain areas without partitioning the channels 216

218 CHAPTER 7. Providing fair services whilst exploiting cell overlap allocated in a cell (and thereby reducing the trunking efficiency) while still ensuring uniform blocking levels within the coverage area. Channel assignment strategies such as in [43] exploited cell overlap, but because they did not take into account the different blocking rates in different areas they were inherently unfair. The UFCA scheme [62] took into account the different blocking rates in each area to make the system fair but it did not exploit CNIR effectively. Based on [62] and [43], a new model has been developed which takes into account the effect of interference on channel allocation based on cell overlap. What UFCA-II effectively does is to take into account the carrier to noise plus interference ratio (CNIR) values, and uses a group of multilevel modulation schemes and multiple channels to achieve better QoS and fairness by exploiting the differences in CNIR that exist between the regions. More specifically, it chooses to pick up a number of channels from the base station that is within range and that has the most available channels. The number of channels required depends on the minimum bits per connection BpC thres threshold (normalised to the frame rate) defined to ensure fair quality of service in all regions and the modulation scheme that can be supported in each channel. Performance Before moving into describing the operation and results of the UFCA-II scheme, it is important to explain in more detail how the model was implemented and what sort of additions / alterations had to be made in comparison to the previous model UFCA. Cell Overlap Based on Minimum Received Power Threshold As mentioned before (Chapter 6 - Section 6.3), cell overlap occurs because of the way the power decreases away from the boresight of the antenna. For UFCA-II, the size of the overlapping area is now determined by setting a minimum received power threshold that is determined from the link budget and is related to the power rolloff with angle from the antenna gain profile [48]. This minimum received power threshold is a function of the transmit power of the base stations on the HAP. So, in effect the radius of the cells is set by the fixed transmit power of the base stations at the platform. This is different from the previous model (UFCA) as cell overlap was entirely based on the actual distance from the centre of the cells. To elaborate further, all the calculations for defining the boundaries of the cells in terms of radius R were made based on a number of assumptions. The first assumption made was that all calculations were made in the absence of interference. Furthermore, the HAP base stations transmit power should be fixed and the value has been calculated using the following equation: 217

219 CHAPTER 7. Providing fair services whilst exploiting cell overlap P TX P G 2 RX = Equation [7.3] TX d G Ref where, P TX is the base station transmit power, d is the distance between the user and the HAP, G TX is the gain of the HAP antennas and G Ref is the reference gain, which can be calculated based on the user antenna gain G RX and the wavelength of the frequency of interest. G Ref G = 4π λ RX 2 Equation [7.4] where λ is the wavelength. The peak directivity D max can be often approximated by [50]: D max = 2 arccos ln ( 2) + 2 arccos n θ n φ Equation [7.5] where n θ and n φ are the indices for optimising directivity at the cell edges [48]. Both are functions of the antenna 3dB beamwidths. The directivity D seen at a point where a user is positioned can be calculated as: max n { ( ( ))} { ( ( ))} θ n cos θ cos φ cos θ sin φ φ D = D Equation [7.6] user user user user where θ user and φ user are the elevation and azimuth angles for our user on the ground relative to the boresight of the base station of interest. Equation [7.7] can therefore be applied to calculate the power received at this position which is a distance R away from where we see the maximum directivity. Thus, the power level required (P RX ) from a user to be able to connect to the base station of interest is defined as follows: P RX GRef = Dx 2 Equation [7.7] d 218

220 CHAPTER 7. Providing fair services whilst exploiting cell overlap Assuming that a user is R units away from the centre of the cell, then let θ x and φ x be the two angles that can be used to calculate the radius of the cell. As a result, the directivity at this position will become D x. Applying Equation [7.6] to Equation [7.7] and then to Equation [7.3] we can calculate the value of the transmit power required from the base station to ensure coverage of a cell of radius R. The following table lists some of the link budget parameters we have used for the simulation: Table 7.4 Default Parameter values used to assess performance No. Parameter Value Unit 1 Noise Received Level (N RX ) dbm 2 Frequency (f) 28 GHz 3 Reference Gain (G Ref ) Platform Height (h) 22 km 5 Transmitter Power (P TX ) dbm 6 Number of Cells (c) 37 7 Channels per Cell Cluster Size (K) 7 9 Cell Radius (R e ) 1 km 10 Overlap Radius (R) 1.25 km 11 Offered Traffic(OT) 1 Erl/km 2 12 RX Antenna Gain (G RX ) 38 db 13 HAP Antenna Gain (G TX ) 39 db 14 RF Bandwidth (BW RF ) 10 MHz 15 Bits per connection per frame (BpC thres ) 15 b/c/f 219

221 CHAPTER 7. Providing fair services whilst exploiting cell overlap Re R Figure 7.8 Definition of Overlap Radii Figure 7.8 depicts a hexagonal layout of cells, and marked on one of the cells are the circle which encloses a hexagon, it has a radius of R e. The value of the overlap radius R varies within limits defined from the minimum power received threshold. The minimum value of the radius of the overlapping cell is chosen to be equal to the original radius of the cell R e in order to avoid leaving any areas without service. As mentioned previously, the maximum value is taken to be equal to 1.5R e radius, in order to limit the maximum number of overlapping cells to three: but theoretically four or even more can overlap if the cell radius is increased sufficiently, assuming the co-channel interference and received power is acceptable. As before, the UFCA-II work has been based on a 6 neighbouring cell model. This is the most common case in a large-scale cellular communication system. The overlap radius R was set to 1.25km whereas the initial radius of the cells R e was set to 1km (Figure 7.8). For an overlapping radius of 1.25km, a fixed transmit power from the HAP of 0.07mW was required (remember that HAP antenna has a gain of 39dB). This value was calculated based on the SNR levels of Table 7.5. Traffic Model Assumptions The following assumptions were made for the simulations that follow. The major difference that can be seen as compared with the UFCA scheme is that user s selection of base stations depends on their minimum received power threshold as opposed to the distance that UFCA employed. Also, new users are blocked if they cause degrading of CNIR levels of existing users. 220

222 CHAPTER 7. Providing fair services whilst exploiting cell overlap 1. Call arrivals are described by a Poisson process with rate λ while the call duration is exponentially distributed with mean 1/µ. 2. Blocked calls are cleared and do not return. 3. Users connect to the base station with the most available channels within range. They can choose channel(s) from one of up to the three base stations depending on the area in which they are located. 4. Both HAP and users are considered to be stationary. 5. Handoff or Call Dropping is not implemented. 6. New users are blocked if they cause degrading of CNIR levels of existing users below the SNR threshold required listed in Table 7.5. Effects of Interference on UFCA scheme performance It has been shown that the UFCA model [62] has been perfectly fair i.e. provides fair channel distribution while ensuring minimum blocking levels. However the data rate was determined by the area with the worst Channel to Noise Ratio (CNR) levels by setting a minimum data rate threshold and having a fixed modulation scheme that could be supported in all areas. This would ensure uniform data rates as well as uniform blocking levels. However, it did not exploit the differences in the CNR levels between the areas. Introducing interference to the new model and using CNIR levels as metric for determining which one of the supported modulation schemes (Table 7.5) to use, it was possible to exploit these differences. To start with, we have first looked at the CNIR plots for the case with no overlap and for the case where we allow overlap of R=1.25km. The following plot illustrates the CNIR levels of the centre cell for cluster size of 7. The full parameter list used can be found in Table

223 CHAPTER 7. Providing fair services whilst exploiting cell overlap Figure 7.9 CNIR levels of No Overlap Model and Overlap Model From Figure 7.9 it can be seen that by allowing overlap of R=1.25km, the users at the edges of the cell (i.e. the users located in the overlap areas) experience lower CNIR levels than the users in the non-overlapping areas (area A). As a result users in these areas cannot connect to high rate modulation schemes that can give higher bit-rates. Although the blocking probability in the areas of overlap is much lower than in area A (and the case of no overlap), connections in these areas will be of much lower data rates. The modulation schemes assumed for this simulation are listed in Table 7.5. UFCA-II Algorithm As mentioned before, the initial UFCA scheme ensured uniform blocking levels between the regions but it has been shown that users in different regions can support different data rates. To overcome this problem the following mechanism in Figure 7.10 has been devised. 222

224 CHAPTER 7. Providing fair services whilst exploiting cell overlap Stage 1 Stage 2 Total Bits/C/F Figure 7.10 Channel allocation flowchart guarantees fair data rate in all areas of a cell Stage 1: To solve the problem, more than one channel is allocated to a user in order to improve the total number of bits per connection per frame (BpC) and therefore increase its data rate to an acceptable level. A user will keep requesting more channels in order to satisfy the minimum acceptable level BpC thres of x bits/c/f (Table 7.5). Channels can support different bits per symbol depending on the modulation scheme. The most extreme case will be that of when all available channels can support only 1 bps (BPSK) and therefore x channels will be required for one user to satisfy the BpC thres. However, better modulation schemes can be supported in the areas with the least interference and therefore fewer channels will be required per user. According to Figure 7.9, users in area A are more likely to request fewer channels than users in area B or area C to reach the minimum bps threshold. Stage 2: Users select the BS within range that has the most available channels. The channel selection is based on which channel is first found to be available in the chosen BS with an acceptable CNIR level for BPSK. The extra channels a user will request should come from the same base station that the initial channel has been assigned from. It could be possible to allow users to be allocated channels 223

225 CHAPTER 7. Providing fair services whilst exploiting cell overlap from any base station within range (for users in area B and C) but for simplicity it was preferred to allow channel assignment from one cell only. However, new users are blocked if they disturb any currently active users by degrading their bit rate level below their current value. In order to prevent this happening too often, a safeguard CNIR margin called Eff-CNIR (Table 7.5) of 5dB has been set when deciding which modulation scheme to use. This value was chosen upon investigation which gave the lowest blocking probability levels. This is based on the concept of hysterics [71] where there is an acceptance Eff-CNIR threshold and an interruption CNIR threshold. By introducing this margin, users are less susceptible to interference and can afford to lose some of their CNIR levels before they actually reach the interruption CNIR levels, thus allowing the new users to use the same channel in a different cell without being block or without degrading the service of existing users. More specifically, new users can connect to the system if their CNIR levels are above the Eff- CNIR levels and the current users CNIR levels do not drop below the interruption CNIR levels. As a result, for this work users connect to the scheme that can be supported from their CNIR level assuming that their CNIR is 5dB less than what it really is. This margin is applied when choosing a modulation scheme from Table 7.5 higher than BPSK, so that users will not be completely blocked as far as their CNIR can support BPSK. In the case that not enough channels have been obtained to sum up and satisfy the BpC thres threshold, all channels that have been reserved will be released and the user will be blocked. The following table lists the modulation schemes that have been used along with their CNIR values. For this work it has been assumed that the interference is Gaussian. Note the standard CNIR value we have assumed and the effective CNIR value that users must fulfil to be able to use the relevant scheme. Table 7.5 Modulation and Coding figures used to determine capacity Mod. Scheme BPSK QPSK 8PSK 32-QAM 64-QAM Bits / Symbol CNIR(dB) Eff-CNIR(dB) The performance of Uniform Fixed Channel Allocation Scheme II (UFCA-II) has been assessed. All tests investigated the effect of CNIR on capacity and blocking levels. The first set 224

226 CHAPTER 7. Providing fair services whilst exploiting cell overlap of results presented is for the case where there is no overlap. Then we introduce overlap but without applying the Random Acceptance Factor (RAF) technique. Finally, the RAF technique is applied. Simulation Model without Overlap Users in this model have to connect to the closest BS, so areas B and C do not exist. The cell radius is set to 1km. More information regarding the parameters used for this simulation can be found in Table 7.4 Figure 7.11 CNIR and Channel usage plots without cell overlap Figure 7.11 depicts the cumulative distribution function (cdf) of CNIR and the cdf of the channels in use for the case of no overlap. These results were used to compare the performance of the no overlap case with the overlap case. Simulation Model with Overlap The radius of each cell now is set to 1.25km. Users can connect to up to 3 BSs depending on their location. This simulation was also based on the parameters used in Table

227 CHAPTER 7. Providing fair services whilst exploiting cell overlap Figure 7.12 CNIR and Channel usage plots with cell overlap From Figure 7.12 it can be seen that we have a multilevel usage of channels in each area. This depends primarily on the CNIR levels which for the overlap areas B and C are worse than for area A. As a result, users in these regions have to use lower modulation schemes and therefore there is higher demand for channels to satisfy the minimum bits per connection threshold level (BpC thres ). Taking as an example the channel usage in area C, we can see that most of the time users require on average more channels than users in area A or B. Although the blocking levels in area A, B and C differs (Table 7.7), we have managed to guarantee uniform bit rate for all users. Applying RAF After ensuring uniform bps service, it is important to ensure uniform blocking service within the coverage area. To do so, the RAF technique is being used. Repeating the same technique as in [62], we have numerically calculated a new set of α and β parameters listen in Table 7.6. After a numerical investigation, it was found that the RAF becomes active when there are only 5 channels left in each of the base stations that are within range. Table 7.6 RAF parameters RAF Parameters α β OT varying OT optimum 6.5 erlang/km 2 6 erlang/km 2 226

228 CHAPTER 7. Providing fair services whilst exploiting cell overlap Figure 7.13 CNIR and Channel usage plots with RAF Figure 7.13 depicts the results for the RAF model. The CNIR statistics vary very little when compared with the previous scheme. However, applying the RAF we can even out the blocking probability between areas A, B and C. Applying the RAF the overall level of blocking is approximately 2.5% compared with the no overlap case with 4.2% as shown in Table 7.7. Clearly, we have improved the QoS significantly at no expense of the capacity of the system. Table 7.7 Blocking levels generated from each scheme Area A Area B Area C No Overlap 4.2% N/A N/A With Overlap 3.4% 1.8% 0.1% UFCA-II 2.5% 2.6% 2.5% Note that overlap areas B and C experience lower CNIRs than area A. As a result, more channels are required to be assigned per connection in these areas in order to guarantee the BpC thres level set. However, users in these areas get increased flexibility that effectively reduces the number of channels a cell may have to assign them. As a result, users in area A will on average have a greater number of channels available. 227

229 CHAPTER 7. Providing fair services whilst exploiting cell overlap 7.7. Summary and Conclusions This Chapter has been dedicated on the exploitation of cell overlap and investigation of various techniques for improving the Quality of Service in terms of blocking probability, data rate and fairness. These techniques are ABFCA, RBFCA, UFCA and UFCA-II. Comparing the ABFCA scheme with the no overlap case showed that the system performance in terms of blocking probability has improved since the overlap areas benefit from the increased channel availability. This however introduces a non-uniform blocking between the areas and therefore across the coverage area that is not fair for the users. To address this problem, RBFCA has been developed. This is an optimisation scheme that allocates channels directly to regions in order to unify the blocking across the cell. From the results, it has been shown that allocating channels into regions improves fairness across the cell but this comes at the cost of much higher blocking levels. Furthermore, it has been shown that trunking efficiency becomes an issue when allocating channels to regions and as a result the optimisation technique cannot guarantee uniform blocking. The results obtain from the UFCA scheme has shown that cell overlap can be exploited to improve the performance for a scheme that uses fixed channel allocation (FCA). The areas served by more than one cell (overlap areas) benefit from the higher trunking efficiency and as a result they have much lower blocking than the areas served by one cell. To improve uniformity and reduce the blocking probability within a cell, a technique has been developed called Random Acceptance Factor (RAF) where certain restrictions have been imposed in order to prevent a proportion of the channels from being allocated to the overlap areas. As a result, these channels are re-directed in areas with no overlap (area A). This technique enabled us to control the number of channels being allocated into certain areas without partitioning the coverage area into smaller regions. It will therefore be of particular benefit in situations where there is nonuniform distribution of traffic and will also aid handoff. The UFCA model [62] has been perfectly fair in terms of the channel distribution while ensuring minimum blocking levels. However the data rate was determined by the area with the worst Channel to Noise Ratio (CNR) levels by setting a minimum data rate threshold of x bps and having a fixed modulation scheme that could be supported in all areas. This ensured uniform data rates as well as uniform blocking levels. However, it did not exploit the differences in the CNR levels between the areas. 228

230 CHAPTER 7. Providing fair services whilst exploiting cell overlap The improved UFCA-II scheme looks at different ways of selecting a base station (BS) and assigning a channel or more to a user. The UFCA-II model exploits both cell overlap and CNIR and delivers uniform QoS. This has been achieved by introducing a bit per connection threshold level BpC thres that all user connections must satisfy. This allowed the CNIR to be better exploited in the channels. This threshold in conjunction with the RAF has ensured uniform blocking across the areas as well as equal data rates across the user connections. For the parameters chosen it has been shown that the blocking probability can be reduced from 4% to 2.6% by exploiting this technique. 229

231 CHAPTER 8. Handoff Techniques and HAP Mobility Models Chapter 8. Handoff Techniques and HAP Mobility Models 8.1 Introduction Handoff Techniques High Altitude Platform Mobility Models Mobility Models and Handoff Steerable Antennas and Handoff for HAP motion correction Handoff Simulation Model Discussion of Results Conclusions Introduction HAPs can achieve a high capacity by using a large number of wireless transceivers, each using a directional antenna to create cells on the ground. As mentioned before, these transceivers are colocated on the platform, and they offer line-of-sight communication to a geographic service area of approximately 60km diameter [9]. It has been shown in [43], [62] and [72] that with the nature of the HAP architecture, it is possible to achieve cell overlap and thus maximise the system capacity whilst ensuring uniform and fair quality of service to all users. Despite the advantages that the HAP architecture has to offer [14], there are also several disadvantages that need to be investigated. One of these disadvantages is the relatively loose station keeping characteristics. Although the stratosphere is a layer of relatively mild turbulence [73], the platform will inevitably encounter sudden wind gusts. As a result, the platform could move in any direction. Various antenna steering mechanism techniques have been proposed that can be employed to practically compensate for the HAP movement. These mechanisms can be used on both customer premises equipment (CPE) as well as on the platform itself. The antenna steering mechanisms [74] [75] on the platform have several constraints associated with them. For example, they must be powered up adding to the power demands on the platform, while they also add extra weight and cost to the payload. Furthermore, even if they are employed it is impossible to guarantee a stationary position for the HAP during its service hours and so the patterns of received power and interference on the ground (footprints) will still be moving. Handoff is a common technique employed by all cellular systems both terrestrial [25] and satellite [76], which has been proven vital both for ensuring uninterrupted connections and 230

232 CHAPTER 8. Handoff Techniques and HAP Mobility Models increasing system capacity (using directed handoff). For this work we will focus on its first use: to maintain uninterrupted connections when one of the ends of the communication link is moving and as a result the user must be transferred from one cell to another (i.e. experience handoff). For terrestrial systems, it is common that the user moves with the Base Station (BS) being fixed. In a HAP system we are assuming that the platform will be moving and the users will be fixed (the case of Broadband Fixed Wireless Access). Taking into account the movements of the HAP, it is possible by employing a combination of handoff techniques and a steering mechanism, to avoid interruptions on the link between the user and the platform. To do so, either the Customer Premises Equipment (CPE) should be able to keep track of the platform and/or the HAP itself should employ an antenna steering mechanism to maintain a constant coverage. As mentioned before, the latter may not be feasible due to the power available onboard and the payload weight constraints imposed by the platform manufacturer. Furthermore, it is important to eliminate unnecessary handoffs / signalling and therefore unnecessary power consumption that could be vital for the constant (24/7) operation of the HAP. There are several handoff techniques proposed in the literature depending on the type of the channel allocation scheme used. These can be classified as fixed [35], flexible [77] and dynamic [78] channel assignment strategies, all initially designed for terrestrial systems. These strategies are used both for new call and handoff call requests. For the case of handoff in a terrestrial communication system, either the user or the base station can make a decision about whether a handoff is required. Both user and base station will monitor the quality of the channel. For this work we have focused on a fixed channel allocation (FCA) based scheme called Area Based FCA (ABFCA) [62] to exploit cell overlap, and have investigated the effect of the various movements that an aerial platform performs in order to quantify the effectiveness of this scheme. The handoff performance is improved when employing a guard channel based scheme. In this scheme a number of channels is reserved explicitly for users requiring handoff [79]. This Chapter starts with examining basic handoff techniques and show the different forms of handoff techniques used in the terrestrial and LEO satellite communication systems. Following, a number of HAP mobility models are presented to draw the difference from the existing terrestrial user movements and the LEO satellite movements. Then the impact that HAP movements have on handoff is presented and it is shown how the handoff can be controlled using mechanical steering correction mechanism. 231

233 CHAPTER 8. Handoff Techniques and HAP Mobility Models An alternative technique to the mechanical steering correction mechanism is then examined, where the handoff simulation model is used to demonstrate the effects of limited mechanical stabilisation for rotation only. Finally, results and conclusions are presented at the end of this chapter Handoff Techniques What is handoff? Handoff is defined as the change of the radio channel used by a wireless terminal. The new channel can be either assigned from the same base station (intra-cell handoff) or from a different base station (inter-cell handoff). In the case where a handoff is unsuccessful and the user is forced to terminate his/her connection, this is called dropping. Handoff can be divided into three types [80]: 1. Handoff Decision / Detection A decision has to be made when exactly to initiate and perform a handoff. Also which cell to handoff to? This decision can be made by the user s equipment or by the base station(s). 2. Handover Resource Assignment This is the area that we will be focusing on. It has to manage the channels in order to ensure that there will be enough channels to minimise dropping probability. 3. Handoff Execution This part of a handoff model includes the protocols for reliable exchange of handover data. This is the signalling procedure needed to inform the handoff call and base station about the new resource allocation. Our investigations focus on the second type of the handoff process. Initial investigations of handoff models designed for terrestrial or satellite purposes have been vital for the development of our HAP-based handoff scheme, and an overview of handoff techniques are presented in this chapter. 232

234 CHAPTER 8. Handoff Techniques and HAP Mobility Models Why handoff is necessary? Handoff is a common technique employed by all cellular systems both terrestrial [25] and satellite systems [76] that has been proven vital for both ensuring uninterrupted connections and increasing system capacity (Directed Handoff). In the first case, a handoff is usually initiated when the radio link drops below a minimum threshold level, and in the second case it occurs when the system is rearranging its channels due to congestion in one area. A typical example for the first case is when a user who is active (i.e. on the phone), moves across the cell boundaries to an adjacent cell (Figure 8.1). In this case the call has to be handed to the neighbouring cell in order to prevent the call being dropped. However, if the new cell does not have adequate channels to support the handoff, then the call is dropped. Channel A Channel B Cell 1 BS1 BS2 Cell 2 Cell Boundary Figure 8.1 Handoff occurs when user moves from one cell to another For the second case where the system is rearranging its channels due to congestion, a number of handoffs take place from congested cells to less congested cells in order to maintain acceptable blocking levels or to further increase the system capacity in a given area [81]. An example of a handoff scheme associated with this concept is the Directed Handoff (DH) scheme [40], [82] and [39]. DH effectively redirects existing calls from one cell to a neighbouring cell to further increase the system capacity. DH requires cell overlap to operate as it redirects existing calls in the overlap region from one cell to a neighbouring cell. 233

235 CHAPTER 8. Handoff Techniques and HAP Mobility Models Quantifying Handoff Handoff performance can be measured in terms of how many handoffs take place in a region or coverage area and how many of these are successful or unsuccessful. To measure the effect of the users or platform movements and assess the performance of the handoff in terms of these two aspects, we consider the Handoff Probability (P H ) and Dropping Probability (P D ) separately. Handoff Probability In this work Handoff Probability is defined as a measure indicating how often users experience a successful handoff irrespectively whether the connection was eventually dropped. P H N N N H = Equation [8.1] B where N H is the total number of users that perform handoff, N B is the number of user that have been blocked and N is the total number of users. So a handoff is required if for example a user moves across the boundaries of a cell (in a terrestrial system) or in the case of a HAP communication system the HAP moves away from the centre of the coverage area causing the same effect on the user. Dropping Probability Dropping probability represents the levels of unsuccessful handoffs in a communication system. Here: P N D D = Equation [8.2] N N B where N D is the number of the users being dropped and N B is the number of users being blocked. N is the total number of users in the system. 234

236 CHAPTER 8. Handoff Techniques and HAP Mobility Models Terrestrial Handoff Schemes In order to minimise the chances of users being dropped during a handoff, a number of channels per cell can be dedicated just for the handoff. These Guard Channel (GC) based schemes [83] [79] provide better performance in terms of dropping probability at the expense of a reduction in the total admitted traffic and an increase in the blocking probability of new calls. Therefore, there is the risk of inefficient spectrum utilisation. Another well-known handoff scheme is the Handoff Queuing (HQ) scheme. The main characteristic of this scheme is that none of the new calls is allocated a channel before the handoff requests in the queue are served. More specifically, a handoff is added to the queue (list of handoffs) when a user has moved to a point where the received power signal from a neighbouring cell is higher than the one the user is currently connected to (point- a in Figure 8.2). When the user s received power signal drops below the minimum received power threshold (point b in Figure 8.2) and no channel has been found, then the call is terminated i.e. is dropped. Figure 8.2 Users are queued for a handoff between point (a) to point (b) Therefore, a call can be queued for a maximum period of time that the user will spend in the region formed by the two boundaries point a to point b. Handoff users are served on a first come first served basis [84] or on a basis of their speed and position in the handoff region [85]. The HQ scheme can reduce the dropping probability at the expense of increased call blocking probability and a decrease in the ratio of carried-to-admitted traffic [85]. Another type of handoff is the New Call Queuing (NCQ) Scheme. This is based on the fact that dropping 235

237 CHAPTER 8. Handoff Techniques and HAP Mobility Models existing users is worse than blocking new users and therefore it is better to queue new calls rather than queue existing calls (like the HQ scheme mentioned before). Results from [83] showed that the dropping probability decreases at a greater rate than the new call queuing blocking probability while the NCQ scheme can support much higher traffic than the guard channel based schemes. On a similar operational basis, the two-handoff-level algorithm [33] employ two handoff request levels as well as a minimum power received threshold level. The way this scheme works is that when the users received power levels drop below the first handoff level, a handoff request is initiated. At this stage a handoff will take place only if the signal from the new cell is greater than the first handoff level of the new cell (case 1 Figure 8.3). Handoff Level - 1 Handoff Level - 2 Handoff Level - 1 Handoff Level - 2 BS - 1 BS - 2 Cell - 1 Cell - 2 Case 1: At this point Level 2 - Cell 2 is greater than Level 1 Cell 1 Figure 8.3 Two Level Handoff Scheme Case 1 [33] In the case where the user reaches the second handoff threshold level, the call will be handed off with no conditions assuming that the user is within the range of the second handoff level of the new cell (case 2 Figure 8.4). In the case where there is a gap (no coverage, case 3, Figure 8.4), or no available channels in the new cell, the call will continue until its signal strength drops below the minimum threshold level. In this case the call will eventually be dropped. 236

238 CHAPTER 8. Handoff Techniques and HAP Mobility Models Handoff Level - 2 Handoff Level - 1 Case 3: This is a hole in the coverage area. Users might be dropped depending on how much time they will be spending in the hole Cell - 1 Handoff Level - 2 Handoff Level - 1 Cell - 2 Case 2: At this point Level 2 - Cell 2 is greater than Level 2 Cell 1 Figure 8.4 Two Level Handoff Scheme Case 2 and 3 [33] The failure of a call to be handed off happens when the required signal quality is below a required value for more than a given time interval (e.g. 5secs - [33]). This time interval depends explicitly on the type of traffic the system supports. If for example the system supports voice calls, then the time interval will be a real-time delay sensitive service that can tolerate only short interruptions. On the other hand, for a non-real-time data service such as internet-based application services, interruption for a longer period of time might be passed unnoticed since only a small number of packets could attempt to be transmitted during that period. Satellite Handoff Schemes Satellite handoff has been largely used in the Lower Earth Orbit (LEO) satellites as they constantly move around the globe in a fixed orbit. Each revolution takes between approximately 90 minutes to few hours, and a group of LEO satellites is employed such that there is always a satellite on a line-of-sight. So, fixed users on the ground might experience inter-cell handoff or intra-cell handoff meaning that a handoff will be required on a cell-to-cell basis or on a satelliteto-satellite basis. Handoff schemes for LEO satellites were designed based on the constant speed repetitive revolutions that LEO satellites perform [76]. Handoff design has been simplified as LEO 237

239 CHAPTER 8. Handoff Techniques and HAP Mobility Models satellite movements are predictable. It is therefore possible to predict the number of cells serving one area at any time High Altitude Platform Mobility Models The HAP may move in any direction at a varying speed and so both the HAP and the users have to cope with these movements. We have looked at all 6 degrees of freedom that a flying object like a HAP (plane or airship) can be subjected to. The movements examined are horizontal displacement with respect to the x, y and z-axis as well as yaw, pitch and roll. In addition the effect of HAP movements has been investigated assuming that the payload is stabilised against rotation but not against drift. In practice the HAP can perform any (or a combination) of the six degrees of freedom as far as it remains within certain boundaries. According to the ITU [18] a platform should be stationed within a circle of radius 400m with height variations of ±700m. According to HeliNet [9] a platform should be stationed within a location cylinder where the cylinder must have a height of 3km height and 4km radius and the HAP must be located within the cylinder for 99% of the time (see Figure 8.5). For this work we have assumed that the platform could either be an airplane or an airship that should be able to serve the nominal coverage area at all times. Therefore, the most appropriate model was the location cylinder model proposed in HeliNet. The radius and the height of the cylinder depicted in Figure 8.5 are defined based on the assumption that the footprint must completely cover the cell area occupied by the users at any time. A HAP communication system will be designed to serve a number of cells on the ground. The number of cells will depend on the type of the terrain that is to be served. For rural areas it is expected that fewer cells of large size will be employed whereas for urban areas more beams of smaller size will be employed. In both cases the HAP is expected to be moving. The movements must be restricted within a nominal volume such that the nominal coverage area is always being serviced. The user antenna is expected to be of high gain (39.4dBi [9]). This sort of antenna will be highly directional and therefore it will be of a particularly small beamwidth. As a result, any movement of the HAP will have a direct impact on the user. The impact will be more significant if the user employs a fixed antenna. This will be a much cheaper solution for the user at the expense of long outages due to the HAP movements. If for example, when the HAP experiences drift and the customer premises equipment (CPE) antenna is fixed then the antenna will be effectively pointing at the sky rather than at the platform. The highly directional CPE antenna becomes blind as it is unable to communicate with the HAP. 238

240 CHAPTER 8. Handoff Techniques and HAP Mobility Models The simplest form of steering to be applied on the platform is an individual antenna correction mechanism that adjusts each antenna such that the boresight points directly to the position on the ground the cell centre should be pointing at. This mechanism however, requires a large number of complex and heavy mechanical gimbals equal to the number of cells required to cover the nominal coverage area. It is therefore not practically feasible. Based on this principle, a four actuator steering concept [74] has been proposed. This is effectively a mechanical mechanism that has been designed to provide correction to a group of antennas independently from another correction that might need to be applied. The four actuator mechanism is presented in more detail in section 8.5 of this Chapter. For this work we have assumed highly directional steerable antennas for the users as well as a stabilised payload against rotation but not against drift. Figure 8.5 HAP position cylinder for 99% and 99.9% of time as defined by HeliNet [9] Although a steering correction mechanism can ensure longer service availability it also adds extra weight to the platform, and also requires electrical power to operate. This implies an additional cost Mobility Models and Handoff The six degrees of freedom presented in Chapter 3, describe all possible movements that an aerial platform may perform. They can be defined in terms of vectors x, y, z, x θ, y θ, z θ where the first three denote movement in a specific direction and the latter three denote rotations with the subscript denoting the axis of rotation. The six degrees of freedom can be grouped as drift and 239

241 CHAPTER 8. Handoff Techniques and HAP Mobility Models rotational-based movements. The drift-based movements are the x, y and z-axis drift movements that a platform might perform. The rotational-based movements are the x, y and z- axis rotational movements; these are also known as pitch, roll and yaw respectively. Simulation of all six degrees of freedom has not been necessary since some of the movements were considered to have the same effect on the system. These are the x and y-axis drift as well as the x and y-axis rotation (roll and pitch). Thus, we have examined the effect of the drift and rotation based on the x and z-axis only. In addition two more complex types of movements called random walk and reflection have been investigated. These two types contain both drift and rotation movements that an aerial platform might perform. Drift with respect to the x, y and z axis x or y-axis drift of the platform have similar effect on the ground. This is because the movement is similar and because x and y-axis are lying on the same plane. Results can therefore represent both cases. Here we have investigated the effect of HAP movements assuming that the payload is stabilised against rotation but not against drift. The HAP is allowed to drift up to a distance of 2.35km away from the centre of the coverage area. This is to ensure that all users are located within the coverage area of the 19 cells. z y x +2.35km -2.35km Figure 8.6 Example of y-axis drift It is also assumed that the HAP will start moving from the centre of the coverage area to one end of the position cylinder and then it will move back to the other end. The HAP does not change its orientation during this movement, to keep each type of movement isolated. We also assume that the current position vector r t is dependent on the vector ( seconds previously. rt t ) generated t 240

242 CHAPTER 8. Handoff Techniques and HAP Mobility Models Mathematically this can be represented as [86]: r = r + v t aˆ' Equation [8.3] t t t where v is the velocity and aˆ ' = a a f f a a f 1 f 1. a f represents the vector corresponding to a destination point A F and a f-1 corresponds to the vector of the point of the previous point update. A F in this example represents the edge of the cylinder that the HAP will move towards. An example of this is shown in the figure below: Final Position (AF) Initial Position (A) a f t(s) t(s) 0 r ( km) â' r 0 ( km) t Figure 8.7 Diagrammatical representation of Equation [8.3] For this mobility model a starting point (0, 0, 17)km is assumed. Then the HAP is set to move to point (0, 2.35km, 17km) and then backwards to (0, -2.35km, 17km). There were no changes in height. Z-axis drift causes expansion and contraction of the size of every cell as well as the whole coverage area. Assuming that there is no steering mechanism on the HAP (apart from the rotation stabilisation mechanism mentioned before), the HAP is allowed to drift up to a distance of 1500m [9] away from its initial point to ensure that all users remain located within the 241

243 CHAPTER 8. Handoff Techniques and HAP Mobility Models coverage area. In both cases where the HAP moves upwards or downwards, the footprints must be able to serve the nominal coverage area. Figure 8.8 Example of z-axis drift More specifically, the footprint must cover the cell area where this is the area on the ground where the loss to one particular HAP antenna is minimum when the HAP is at the centre of its coverage and it is not moving (i.e. initial position). In this case, the size of the footprint will change whilst the HAP moves upwards or downwards. However, the effect depends on the antenna mask employed. In the case where the HAP moves downwards (point A Figure 8.9), the footprint could get smaller provided that the antenna profile looked like the one in Figure 8.9 B. On the other hand, if the antenna profile used was the one in Figure 8.9 A, then the effective size of the footprint might become bigger since the platform is closer to the ground. Although users in this case might be at a much greater angle from the boresight (e.g. user A example in Figure 8.10), the R 2 is smaller and this might put the user within the footprint. 242

244 CHAPTER 8. Handoff Techniques and HAP Mobility Models A. Less Directional B.Highly directional β1 β2 Figure 8.9 Antenna mask example However, if Figure 8.9 B profile is used, then the footprint will shrink and the user A illustrated in Figure 8.10 will not be within range. On the other hand, when the HAP moves upwards things are different. In this case, when employing the antenna profile B of Figure 8.9, the footprint range will increase (see Figure 8.10). Thus the users that were located outside the footprint when the HAP was at its initial state will now have a smaller angle with respect to the boresight. As a result it is possible that they could actually connect to this antenna. If however Figure 8.9 A profile is used, then when the HAP moves upwards, the power density on the ground is expected to be lower (as the antenna is less directional) and as a result the footprint range will be reduced. 243

245 CHAPTER 8. Handoff Techniques and HAP Mobility Models High Position A. As the HAP moves downwards, the power density on the ground increases. As a result the size of the footprint boundaries on the ground shrink. Initial Position Low Position Initial Position B. As the HAP moves upwards, the power density on the ground drops significantly. As a result the size of the footprint boundaries on the ground shrink. User A User B Cell Initial Footprint Cell Initial Footprint Figure 8.10 Drift on the z-axis scenario It is therefore difficult to describe the effect using an equation since it is essential to know the antenna mask used. Rotation with respect to the x, y and z axis Pitch and Roll have similar effect on the ground. This is again because the rotation is performed on two axes both lying on the same plane. In both cases the HAP rotates +/- θ degrees such that there are no users left without coverage. For this work we consider that θ varies between: R arctan fprint - R h cell xθ R arctan fprint - R h cell Equation [8.4] R arctan fprint - R h cell yθ R arctan fprint - R h cell Equation [8.5] 244

246 CHAPTER 8. Handoff Techniques and HAP Mobility Models z y x θ Figure 8.11 Example of y-axis rotation (pitch) For the case of the rotation with respect to the z-axis i.e. yaw, we consider that the HAP rotates at a constant speed anticlockwise (viewing it from top-view) as shown in the figure below. z y x θ Figure 8.12 Example of z-axis rotation (yaw) Planes such as Unmanned Aerial Vehicles (UAVs) are expected to be in a constant rotational type of orbit with respect to centre of the coverage area in order to maintain service over the coverage area. Random Walk To examine a more realistic type of movement, drift and yaw have been combined together to give what we call random walk. In this case the HAP moves at a constant speed in a random direction. The HAP must employ some sort of propulsion mechanism to keep it within the cylindrical boundaries as defined in HeliNet [56]. 245

247 CHAPTER 8. Handoff Techniques and HAP Mobility Models End Start (A) Figure 8.13 Example of random walk HAP movement Figure 8.13 illustrates an example of the route that a HAP takes when assuming random walk and the movement can be described in terms of horizontal and vertical components. The direction changes after a fixed time period t during which the HAP also experiences drift. The greater the speed v h of the platform the longer the distance it will travel within the time period t. After the HAP starts from position A, it is assumed that the HAP continues on the same direction (vector) over the update interval t, with velocity v h. The new horizontal component of position is therefore anywhere on a circle of radius r based on the HeliNet cylinder [9] at a randomly distributed angle θ. The decent/ascent rate is assumed to be uniformly distributed between +/-v v. Therefore the new locus of the new position A' is an ellipsoid centred on point A (assuming it does not hit the edge of the position envelope). Mathematically this can be represented as [86]: r t [ v ( θ ) i + v sin( ) j v k] = r + t θ + t t h cos Equation [8.6] h v Reflection For the reflection type of movement the HAP moves from its current location to a new location at a predefined speed. The direction and the distance are randomly selected. The HAP position is maintained within the cylindrical boundaries defined in HeliNet. 246

248 CHAPTER 8. Handoff Techniques and HAP Mobility Models End Start Figure 8.14 Example of reflection HAP movement Based on the same terminology used in the previous example also illustrated in Figure 8.7, a point A F is randomly selected anywhere within the position envelope according to a random uniform distribution. This represents the destination point the HAP is heading towards at a constant speed v. The intermediate positions are calculated every time interval t based on the speed of the platform. When the HAP reaches its final destination A F, the process is repeated [86]. r = r + v t a' Equation [8.7] t t t ^ where a a ^ a' = f f 1, with a f representing the vector corresponding to point A F and a F-1 f a a f 1 corresponding to the vector of the point of the previous point update Steerable Antennas and Handoff for HAP motion correction Steerable antennas can be used to cope with the movements of the HAP. They ensure constant coverage at the expense of increased weight of the payload if mechanically steerable antennas are employed. Operating at 38-40GHz bands means that for the time being electrically steerable phase array antennas are not practical. Thus corrugated horn antennas [52] with exceptionally low sidelobes are assumed. 247

249 CHAPTER 8. Handoff Techniques and HAP Mobility Models HAP movements have been addressed in the past such as in [74], [87] and in [75] where various techniques have been proposed to cope with various movements. In [75] a propulsion mechanism is proposed in order to counterbalance the horizontal displacement with the ideal position of the HAP and the relevant correction required being specified using a global positioning system (GPS). In [87] it has been proposed that for the inclination effect, a gimbaling mechanism can be used at the bottom of the platform. This will limit the movement of the antennas with respect to the ground. In [74] a steerable antenna correction mechanism different from the previous ones was proposed. This mechanism was developed in the University of York. In this work it was shown that the best steerable antenna mechanism was the one that was applied on every antenna individually. This was however not considered to be an ideal solution as it would require a complex mechanical system with a large number of motors and therefore it would add significant weight to the payload. It was therefore preferable that a HAP system would employ some sort of mechanically steerable mechanism but for a group of antennas instead. This mechanism is presented in more detail in this section. Aperture Antenna Steering Solutions [74] In Chapter 3, it has been shown that HAP movements can change position or distort the shape of the individual cells. In the case where the HAP drifts from the centre of the coverage area, cells move from their intended position. Also, when the HAP experiences roll or pitch a part of the coverage area might be left with no service. Also, the individual cell shape is distorted in terms of the received power as the angle of incidence on the ground of the main beam changes due to the roll or pitch. The power levels form an approximately elliptical footprint instead of the indented approximately circular one (see example in Figure 3.3 page 49). Assuming a steerable antenna grid mechanism, the HAP antennas employed can be redirected (steered back as a group and not individually) to point towards the centre of the coverage area [74]. 248

250 CHAPTER 8. Handoff Techniques and HAP Mobility Models New Position Drift Initial Position Steerable antennas can be used to maintain constant coverage area. New SPP SPP Coverage Area Figure 8.15 Steerable antennas can be utilised to correct drift movements and therefore maintain constant coverage area. What will happen in this case is that the HAP antennas will point to a different location (except the centre cell) than the one they did when the HAP was located at the centre of the coverage area. In addition, all antennas will have a different elevation angle. This means that the power distribution of the ground will not be the same (compared with the initial case where SPP was at the centre of the coverage area). Nevertheless the HAP will still be servicing the intended coverage area thus correcting the drift motion. The example shown in Figure 8.15 depicts the HAP when drifting along the x and y-axis. However, HAP can also drift on the z-axis causing the footprint area to expand or shrink depending whether it goes upwards or downwards. Initial Position HAP normal height New Position HAP moves Downwards HAP moves Upwards SPP Coverage Area Figure 8.16 Steerable antennas can be utilised to correct drift movements with respect to the z-axis. 249

251 CHAPTER 8. Handoff Techniques and HAP Mobility Models Figure 8.16 depicts a mechanism that was build at the University of York [74] that could potentially be used in order to cope with drift on the z-axis. The way this mechanism works is that all rings are interconnected so that when the HAP moves upwards or downwards, the cells of each ring will change their pointing angle symmetrically (except the centre cell which will keep pointing at the centre). Centre Cell 1 st Ring 2 nd Ring 3 rd Ring 3 rd Ring 2 nd Ring 1 st Ring Centre Cell HAP moves Downwards HAP moves Downwards Figure 8.17 Steering mechanism for correcting HAP drift on the z-axis If for example the HAP moves downwards as shown in Figure 8.17, the 1 st ring to the N th ring will set to look outwards whilst the centre cell will be pushed to move slightly downwards (due to the mechanism). The cells will therefore continue to point to the same locations but with a different elevation angle. As for the centre cell, its coverage will shrink unless a correction lens is employed to dynamically correct its beamwidth according to the new height. In the case of roll or pitch the HAP s sub platform point (SPP) remains the same but the antennas are pointing in a different direction. The antennas can be steered back to their original position by employing the mechanism described below for the case of drift in the x or y-axis. 250

252 CHAPTER 8. Handoff Techniques and HAP Mobility Models New Position Initial Position Steering Mechanism Grid of Corrugated Horn Antennas SPP Coverage Area Figure 8.18 Steerable antennas can be utilised to correct pitch and roll movements and therefore maintain constant coverage area. In this case, the antennas will still be pointing towards their intended positions and therefore the elevation angle with respect to their boresight will not be affected. The steering mechanism must be intelligent enough to cope with a combination of these movements. Steerable antennas can be employed to guarantee constant coverage at all times assuming that the HAP is located within certain boundaries Handoff Simulation Model To determine whether the additional expense of carrying a mechanically steerable antenna correction mechanism is necessary, it is important to investigate whether the platform could possibly operate without any major correction mechanism (by just being stabilised against rotation but not against drift). Therefore, it is necessary to examine the effect of the movements of the platform first and see whether it is possible by employing other techniques such as handoff to overcome the problem of station keeping. Since a HAP communication system is a centralised system it could be practically possible to perform all these necessary computations for the handoffs on a central ground station (backhaul) which then feeds all the data to the HAP which then bounces off to all the users. This is of course assuming that we have a ground station that connects the HAP with the global network. Some cell overlap is inevitable because of the way the power decreases away from the boresight of the antenna which has also been exploited to improve capacity as shown in earlier chapters. This is determined from the link budget and is related to the power rolloff with angle from the 251

253 CHAPTER 8. Handoff Techniques and HAP Mobility Models antenna gain profile. Detailed information regarding cell overlap can be found in Chapter 6, Section 6.3. For this work we have assumed 19 cells of 3.15km radius each. Users have been uniformly randomly distributed within a circle of radius 9km and the HAP was allowed to move within limits such that the users would not be left without service. The following figure illustrates how users are being distributed and how the footprints are positioned on the ground km 9km Figure 8.19 Users are generated within a circle of 9km and served by 19 cells of 3.15km radius each 252

254 CHAPTER 8. Handoff Techniques and HAP Mobility Models The following table lists the general system parameters for this work. Table 8.8 Handoff Scenario Simulation Parameters No. Parameter Value Unit 1 Noise Received Level (N RX ) dbm 2 Frequency (f) 28 GHz 3 Reference Gain (G Ref ) -90 db 4 Platform Height (h) 17 km 5 Transmitter Power (P TX ) dbm 6 Number of Cells (c) 19 7 Channels per Cell Total Offered Traffic (OT) Erlangs 9 Cluster Size (K) 7 10 HAP height (h) 17 km 11 Cell Radius (R) 3.15 km 12 RX Antenna Gain (G RX ) 31.3 dbi 13 HAP Antenna Gain (G TX ) 39 dbi 14 RF Bandwidth (BW RF ) 10 MHz 15 HAP Initial Position (0,0,17) km 16 User Area Radius (R User ) 9 km 17 HAP Speed (V HAP ) km/h The simulation performed was based on a Monte Carlo Simulation model. After every small time interval, the simulator checked whether the position of the HAP had changed and then initiated handoff if required. Figure 8.20 illustrates a general state flow diagram of the handoff algorithm performed. 253

255 CHAPTER 8. Handoff Techniques and HAP Mobility Models HAP New Position Check Position Different Same No Handoff Required Identify and Remove Handoff Users* Add Handoff Users* Update HAP Position * further expanded Figure 8.20 Immediate Handoff Initiation Algorithm If the position of the platform has changed, then all users that have been affected must first be identified and be removed temporarily from the system. The users must then be added back into the system and connected to the new cell. This is to eliminate the case where users are being dropped from a cell that is waiting for some of its current users to be connected to another cell. In this case the cell will have a number of channels available as soon as its handoff users release the channels they occupy. The point is to ensure that these channels are available for the new handoff users coming to the cell. This simulation was performed when allowing users to connect to the closest base station (case 1 No Overlap) or when allowing users to connect to any base station within range (case 2 With Overlap). In the 2 nd case where cell overlap was allowed, the users were potentially able to connect to a number of BSs depending on their received power and the minimum received power threshold that defined the boundaries of a cell. Furthermore, the users would connect to the BS that had the most available channels. Details on how to calculate the minimum received power threshold can be found in Chapter 7 (Section 7.6 page 218) as well as in [72] in Appendix - A.3 (page 290). Any user whose received power level P RX drops below the minimum received power threshold P TH requires an immediate handoff. Figure 8.21 illustrates the state flow diagram for identifying and removing the users from the system. Users that have been removed will then have their state changed to try again. The 254

256 CHAPTER 8. Handoff Techniques and HAP Mobility Models code goes through the trying again users in numerical order to attempt to connect them to a new cell. The handoff will be successful or unsuccessful, depending on whether the new cell has any channels available. In the case where cells overlap with each other, the users will connect to the cell with the most available channels. HAP New Position HAP Previous Position Check Position Same No Handoff Required Different For N users Identify and Remove Handoff Users Next User (n) no PRX<PTHR yes Release Channel User State: Try Again ++n == N no User Continues User State: ON yes Add Handoff Users* Update HAP Position * further expanded Figure 8.21 Identify and Remove Handoff Users Figure 8.22 shows that all users lined up for a handoff must first choose which cell to connect to. As mentioned in [62] and [72] the BS is selected based on the number of channels available. However this is not restrictive and it could be that the users could connect to the cell that offers the best received power or even CNIR. Nevertheless choosing the cell with the most available channels improves fairness within the system [72]. 255

257 CHAPTER 8. Handoff Techniques and HAP Mobility Models HAP New Position HAP Previous Position Check Position Same No Handoff Required Different Identify and Remove Handoff Users* Next User (n) no n = N no For N users User State: Try Again yes Choose a Cell within range Call Dropped State: Dropped no Channel Available yes Call Handoff State: On yes Update HAP Position Add Handoff Users Figure 8.22 Add Handoff Users In the case of no overlap, users always connect to the closest virtual base station. In this case the immediate handoff initiation algorithm illustrated in Figure 8.22 is repeated, with the difference that the users are not identified based on their received power but on their distance from the virtual base station. This is based on the assumption that users will not be connecting to the virtual base station with the lowest loss but the ones that are physically closest. Thus, to identify (and remove) the handoff users we use their distance from all virtual base stations (after the HAP has moved). Also, when adding the user back to the system, the cell is chosen based on the distance of the user from the centre of the cell. In some of the simulations, in order to improve the dropping probability a number of channels have been reserved from being allocated to the new users. Thus, these channels are dedicated to the handoffs. More specifically 10% of the total number of channels have been reserved for handoff in order to provide sufficient flexibility for handling handoff without causing significant blocking Discussion of Results The simulation performed was based on the mobility models presented previously. To give a reasonable simulation time, a set of 19 circular cells was used. A cluster size (K) of 7 was employed and each cell had a group of 100 channels within its coverage area. We have also 256

258 CHAPTER 8. Handoff Techniques and HAP Mobility Models assumed that there is a direct line-of-sight communication between the user and the HAP. The traffic was set to be uniformly randomly distributed within the coverage area of radius R user. The total Offered Traffic (OT) in the 19-cell coverage area was 742 Erlangs and the conversation length was set to 5 minutes. When cell overlap was permitted it was set at 25%, thus the new cell radius would be 25% bigger than before. All simulation parameters are listed in Table 8.8. Statistics are gather from the centre cell except the case where the handoff and dropping probability are presented with respect to the distance from the centre of the coverage area. Drift Movement Results For the x-axis drift movement the platform was set to move on a straight line on the x-axis from one side of the position cylinder to the other starting at the centre. The speed varied from 0 to 200 km/h. Figure 8.23 Handoff and Dropping probability for drift movement Figure 8.20 illustrates that the handoff probability increases in all three cases. We can also see that the handoff probability for the overlap case is slightly higher than the other two. This is because the increased perimeter of the cell boundaries affects more users while the HAP is moving. Dropping occurs primarily because of the lack of channel availability (in a HAP scenario). Nevertheless, random peaks in the spatial location of active users could pose a more significant problem in a HAP communication system. This occurs when users were initially served across two cells are now served by only in one cell (Figure 8.24). It is therefore difficult to guarantee 257

259 CHAPTER 8. Handoff Techniques and HAP Mobility Models zero dropping probability as the random distribution will give rise to local user hotspots which when connections are admitted may fall across into two footprints (Figure Case 1). However, as a result of the platform movements after some time there may only be one single footprint (Figure Case 2) covering the cell (user hotspot area), which alone may have insufficient capacity to support all these users resulting in dropping. Case 1: Two footprints serve one cell Footprint 1 Footprint 2 HAP MOVE Cell (users hotspot) Footprint 1 Footprint 2 Footprint 3 Cell (users hotspot) Case 2: One footprint serve one cell Figure 8.24 Moving fotpints cause dropping For the case of drift movement, the dropping probability has been significantly reduced when applying a 10% Guard Channel mechanism. Furthermore, the overlap case has a significantly lower dropping probability than the no overlap case. It is also noticeable that the dropping probability for the case with no overlap, when HAP drifts at high speeds, has decreased. This is because at 200km/h users are subjected to successive handoffs which within that time the spatial location of users does not change. For high speed the traversing across the cells happens more frequently in comparison with the activation or clear down of users. So, there might be a situation where users are actually undergoing multiple handoffs. If they can successfully undergo one handoff it means that the spatial density of the users is fine at that moment in the time. I.e. after the first handoff situation the user density of active users is acceptable for a single cell to be able to cope with it. The dropping in this situation is lower in comparison with the total number of users supported. 258

260 CHAPTER 8. Handoff Techniques and HAP Mobility Models A B C Distance with no overlap: Figure 8.25 Cell Boundaries with no overlap Figure 8.25 illustrates the boundaries of the rings of footprints that were used to quantify the level of handoff with respect to the distance from the centre of the coverage area when cell overlap was not considered. Figure 8.26 illustrates the boundaries for the case where overlap is allowed and is set to be at 25% (i.e. Overlap radius R is 25% bigger than the cell radius R e.). Some of the results were based on these two diagrams. Since the users are fixed and the HAP is moving, the footprints will be moving along the x-axis (or y axis). Distance with overlap: 0km 3.94km: 9.4km: 1.52km: 6.98km: Figure 8.26 Cell Boundaries with overlap Figure 8.27 illustrates the handoff and dropping probability with respect to the distance from the centre of the coverage area at a constant speed of 100km/h. More specifically, the handoff and 259

261 CHAPTER 8. Handoff Techniques and HAP Mobility Models dropping probability are calculated based on a number of concentric rings with thickness of 500m. Results indicate that the handoff increases as we move away from the centre of the coverage area. In addition it can be seen (for the No Overlap case) that users that are closer to the boundaries of the cells (see point A and B in Figure 8.25) experience a higher dropping probability because they are more likely to require handoff. With Overlap No Overlap No Overlap and 10% Channel guard No Overlap No Overlap and 10% Channel guard With Overlap Figure 8.27 Handoff and Dropping probability with respect to the distance from centre of coverage area However, in the case where we employ cell overlap, the dropping probability has decreased significantly. Increasing the cell boundaries by 25% and allowing users to connect to any of the virtual base stations within range we have effectively increased the trunking efficiency [62] and the ability of the system to cope with a higher spatial density of users. Since handoff users are the users that are at the edge of a footprint, this means that these users are most certainly located within an area of overlap. Thus they can benefit from the greater channel choice. As a result, handoff users have more chances of getting a channel than before (with no overlap) and therefore less chances of being dropped. The 10% guard channel mechanism has performed a lower dropping probability but this is at the expense of much higher dropping probability. Here it is important to say that the period of oscillation at 100km/h has been approximately 10 minutes whereas the mean call length assumed was 5 minutes. As a result users will be subjected to handoff depending on their location and the time their connection has started. 260

262 CHAPTER 8. Handoff Techniques and HAP Mobility Models Figure 8.28 Blocking Probability Examining the blocking levels presented in Figure 8.28, it can be clearly seen that the blocking probability is significantly higher in the case of the 10% guard channel scheme. The blocking has been tripled in comparison to the no overlap case. Furthermore, the blocking probability for the non-overlap case has decreased as the speed increased. This is due to the higher dropping levels experienced at higher speeds. There is therefore a better channel availability for the new incoming users. The overlap case performs better than the other two in terms of the blocking probability. More specifically the blocking probability has decreased from 6.5% to 2% compared with the non-overlapping case. Pitch Movements For the pitch movement the platform was set to perform a uniform angular variation with constant speed. The HAP is assumed to be positioned above the centre of the coverage area so that its sub platform point (SPP) remains constant. The angular velocity was varied so that the speed experienced at the centre of the coverage area varied from 0 to 200 km/h. From Figure 8.8 it was assumed that the centre of the coverage could only be displaced by dkm. Thus the maximum possible value for x θ can be expressed from: d xθ Max = arctan Equation [8.8] h The motion is set to start from idle, and then move to one side (x θmax ) around to the y-axis and then move back on the other side (again to an angle x θmax ). x θmax was set to be equal to 0.16 radians assuming that d was 2.73km, in order for the footprint to keep providing coverage to the cell. 261

263 CHAPTER 8. Handoff Techniques and HAP Mobility Models z x y h xθmax xθmax Figure 8.29 Pitch HAP movement in three stages Figure 8.29 illustrates the HAP movement as has been described before. Notice that the cell area is always positioned within the coverage area of the footprint. 262

264 CHAPTER 8. Handoff Techniques and HAP Mobility Models No Overlap With Overlap No Overlap With Overlap Figure 8.30 Handoff and Dropping probability for pitch HAP movement Figure 8.30 illustrates that the handoff probability increases with or without overlap. We can see that the handoff probability for the overlap case is higher than the case with no overlap. This is again because the increased perimeter of the cell boundaries affects more users while the HAP is moving. However, the dropping probability for the overlap case is significantly lower than the case with no overlap. This is again because users requiring handoff are located in an overlap area near the edge of the cell they are connected to. Thus they benefit from the high trunking efficiency provided in the region due to the overlap. Figure 8.31 illustrates the handoff and dropping probability with respect to the distance from the centre of the coverage area (see Figure 8.25) at a constant speed of 100km/h. The handoff probability is found to be higher for the case with overlap. Looking at the overlap case, a peak at about 5.3km indicates that the users located in the 1 st ring experience high handoff probability. This is because the oscillating type of movement changes the angle of incident of the footprints on the ground. As a result their shape is not circular anymore but approximately elliptical. 263

265 CHAPTER 8. Handoff Techniques and HAP Mobility Models No Overlap With Overlap No Overlap With Overlap Figure 8.31 Handoff and Dropping probability with respect to the distance from centre of coverage area The blocking probability results have shown that the case with no overlap suffers from higher blocking levels. Also, the phenomenon of reduced blocking probability at higher speeds (no overlap case) is as a result of the increased dropping probability, which therefore allows more new users to join the system. No Overlap With Overlap Figure 8.32 Blocking Probability Reflection, Rotation and Random Walk Movement Results Reflection, Rotation and Random Walk have been examined for Handoff and dropping probability performance. In the case of the Rotation mobility model, the rotational movement has been quantified in terms of the two rings of cells that form the coverage area (Figure 8.25). 264

266 CHAPTER 8. Handoff Techniques and HAP Mobility Models Based on Figure 8.5, all HAP movements allowed (applicable in the cases of the Reflection and Random Walk mobility models) were up to 2.36km away from the centre of the coverage area. That is to ensure uninterrupted coverage for all users. Figure 8.33 illustrates the handoff and dropping probability for all three mobility models in terms of speed (km/h). For the Reflection model, the handoff probability increases at a higher rate when HAP moves at low speeds and then it starts to saturate reaching a Handoff probability of 70% at 200km/h. For the case of the random walk, the rate of the increase in Handoff probability is approximately constant at 4% per 10km/h up to 100km/h and then drops to approximately 2% per 10km/h reaching a maximum Handoff probability rate of 61% at 200km/h. Figure 8.33 Handoff and Dropping probability for a HAP system as a function of ground speed when subject to different types of motion For the case of the rotation mobility model, the rotational velocity was set with respect to the centre of one of the cells in the 2nd ring (this is the external ring in a 19-cell model). Thus the speed of the cells was proportional to the distance from the centre of the coverage area and the rotational velocity set. From the results illustrated in Figure 8.33 it can be seen that in both rings of cells the users experienced the same level of Handoff probability irrespectively of their position. The rate of change of Handoff probability with respect to the HAP rotational velocity is 1.8% per 10km/h. 265

267 CHAPTER 8. Handoff Techniques and HAP Mobility Models For the case of the Dropping probability it can be noted (Figure 8.33) that the results behave similarly to the Handoff probability results. For the Reflection mobility model, the Dropping probability increases at a higher rate when having lower speeds reaching the level of 1.5% Dropping probability at maximum speed of 200km/h. The Random walk increases in a more linear manner reaching a point of 1.8% Dropping probability. The results for the Rotation model show that some of the users in the 1st ring will be dropped whereas none of the users in the 2nd ring will be dropped at all. This is because the number of users located in the second ring is much smaller than the users located in the 1st ring since the coverage area is set to be bigger than the area the users are located in. The effect on Handoff and Dropping probability has also been examined as a function of the distance from the centre of the coverage area. Figure 8.34 illustrates the behaviour of all three mobility models in terms of this distance. For the case of the reflection mobility model, the further the users are located from the centre of the coverage the higher the handoff probability they will experience. This is because the HAP is set to drift from one site to the other of the HeliNet HAP location cylinder within a time interval during which users are greatly affected. It can also be said that the handoff probability decreases slightly at the centre of the 1st ring. This is because some of the users are centrally located within a cell (of the 1st ring). Therefore, they have less chances of being handed off since they are away from the boundaries of the cells they are connected to. The Handoff probability decreases as we move to the second ring to become zero at 9km, which is the distance of the most distant user. Random Walk mobility model results had a similar behaviour to the reflection mobility model. However the results showed lower handoff probability levels except at the edge of the user positional area (most distant users at around 9km form the centre of the coverage area). The rotation mobility model showed an even smaller handoff probability. Again, the handoff probability increases the more we move away from the centre of the coverage area. 266

268 CHAPTER 8. Handoff Techniques and HAP Mobility Models Figure 8.34 Handoff and Dropping probability with respect to the distance from the centre of the coverage area For the dropping probability the reflection mobility model is higher than the rest. However for both reflection and random walk mobility models the dropping probability increases the more we move towards the edges of the cells. For example the edges of the cells of the centre ring in Figure The reason the users located towards the outside of the coverage area experience lower dropping than the users located towards the centre of the coverage area is because some of the time these users are connected to the cells of the 2 nd ring that are less congested. The Rotational movement however shows a constant increase of the dropping probability as we move away from the centre of the coverage area up to the point of the most distant user Conclusions In this Chapter, an investigation of the impact of the aerial platform movements in a HAP telecommunication system has been carried out. Results have shown that a handoff technique reduces the need for mechanical stabilisation. It has also been shown that handoff mechanism is necessary to ensure continuity of the connections being affected by the platform movements. Furthermore, by employing guard channels we have ensured significantly lower dropping levels but at the expense of higher blocking levels. More specifically the dropping probability was reduced from 0.2% to almost 0% when compared with the no overlap case. The immediate handoff scheme based on ABFCA cell overlap model has shown that by allowing overlap, the blocking probability has decreased from 6.5% to 2% when compared with the 10% Guard 267

269 CHAPTER 8. Handoff Techniques and HAP Mobility Models Channel case in the x-axis drift case scenario. It was also shown that the dropping probability has been significantly reduced (when compared with no overlap case) since the users that required handoff where located in the areas of overlap. Thus, they experienced higher trunking efficiency and as a result the handoff was successful. Additionally three complex types of motion have been investigated. These are the rotation, reflection and random walk type of movement. This effect is worse for rotation type motion, with the probability of dropping dependent on the cell dwell time. Ways to improve performance is by either reserving channels for future handoff use or controlling the density of admitted users within one area [86]. Hence, the limitation on performance of the HAP system will therefore depend primarily on the resource allocation process, and availability of resources on the platform. Employing cell overlap has once more shown that the quality of service was significantly improved since both blocking and dropping probability were significantly reduced. From this it could possibly be concluded that cell overlap in a combination with UFCAII scheme can provide a uniform low blocking and low dropping performance unlike the other schemes examined. 268

270 CHAPTER 9. Summary and Conclusions Chapter 9. Summary and Conclusions 9.1 Summary and Conclusions Novel Contributions Future Work Summary and Conclusions This thesis explores the use of cell overlap to improve capacity and quality of service (QoS) in a High Altitude Platform (HAP) communication system environment. Analysis and extensive Monte Carlo simulation is used to evaluate performance, focusing especially on the impact of the HAP movement on the communications. Below a brief summary and conclusions for the entire thesis is presented; more detailed conclusions are provided at the end of each chapter. Starting with Chapter 1, this chapter provides the reader with an overview of the terrestrial communication systems, the fixed wireless access (FWA) infrastructure and the HAP architecture but most importantly it outlines the hypothesis of this thesis. Following in Chapter 2, the reader can read through the basics of the Resource Allocation Techniques that are employed in wireless communications. Along the way, an extensive series of computer models have been developed, and these were based on a core simulation model. Chapter 3, 4 and 5 in this thesis are dedicated to the design, implementation and verification of the core simulation model that was extensively used in this thesis. From the results in Chapter 5, it has been concluded that a HAP system can deliver a symmetrical hexagonal cellular layout. Changes in the clustering or directivity of the antennas had a direct impact on the size and shape of the cells on the ground as being defined from their boundaries (minimum received power levels Chapter 3 Section 3.6). The advantage of the symmetrical cellular layout of the HAP on the ground is being stressed through the concept of elliptical beams. Furthermore, it has been shown that the movements of the HAP can cause significant changes to the received power levels at the user s end on the ground that is significantly affecting the coverage in some regions. In Chapter 6, the concept of cell overlap has been introduced and analysed. Based on the cell overlap, a number of advanced fixed channel allocation FCA-based schemes have been 269

271 CHAPTER 9. Summary and Conclusions investigated called Channel Allocation with Cell Overlap Support (CACOS) and Non Overlap Area Support (NOAS). From these schemes it was shown that the cell overlap improves the QoS in terms of the overall blocking probability but nevertheless causes uneven blocking across the different types of areas of overlap. To try and maintain a fair service (uniform blocking), a channel allocation technique has been proposed called Numerical Calculation for Uniform Blocking (NCUB) that takes into account the total number of channels available, the offered traffic (OT) and the cell overlap in order to perform the channel allocation. It was shown that this scheme improved uniformity in terms of blocking but its performance depended on the number of channels available per cell. The distribution of the channels in a cell was then further analysed to investigate the sensitivity of the system in terms of blocking probability for different levels of trunking efficiency. Finally, NCUB was examined under different system parameters such as different size of overlap or different OT. A more complex HAP communication system simulator has been employed in Chapter 7. This was a Monte Carlo based simulation model developed based on the concept of ErlangB traffic model. Using this simulator a number of channel allocation schemes have been presented. The first scheme examined was the simple case of a FCA scheme with no cell overlap. The results from this scheme were used for comparison with the Area Based FCA (ABFCA), Region Based FCA (RBFCA), Uniform FCA (UFCA) and Uniform FCA II (UFCA-II) schemes proposed in Chapter 7. ABFCA showed that areas served by more than one cell (overlap areas), benefit from the higher trunking efficiency and as a result they have much lower blocking than the areas served by one cell. The performance of ABFCA was better than the simple case of FCA with no cell overlap for all three types of areas (A, B and C). However, the blocking levels across the different areas forming a cell were different. RBFCA showed that uniformity in terms of blocking probability across the different regions is achievable. This was done by allocating channels on a regional basis rather than on an area or cell basis. Nevertheless, dividing channels into smaller regions put constraints on flexibility and decreased the trunking efficiency. UFCA has been developed to improve uniformity and reduce the blocking probability within a cell. Unlike RBFCA, channels were allocated on a cell basis rather than on a regional basis. However certain restrictions have been imposed in order to prevent a proportion of the channels from being allocated to the overlap areas. As a result, these channels were re-directed in areas with no overlap (area A). Simulation results showed that the centre cell can now cope with approximately 10.5% more offered traffic (OT) than in the standard case with no overlap. This improvement does not require any prior knowledge of the interference environment. This technique enabled us to control the amount of channels being allocated into certain areas 270

272 CHAPTER 9. Summary and Conclusions without partitioning the coverage area into smaller regions like RBFCA. It will therefore be of particular benefit in situations where there is non-uniform distribution of traffic and will also aid handoff. The UFCA model [62] has been perfectly fair in terms of the channel distribution while ensuring minimum blocking levels. However the data rate was determined by the area with the worse Channel to Noise Ratio (CNR) levels. Thus users in the overlap areas would experience lower data rates. In order to ensure a uniform data rate to the users UFCA-II has been proposed. UFCA-II is effectively based on UFCA scheme but this time a minimum data rate threshold of x bps has been set for the users to meet. Users with in with poor CNIR levels (and therefore low data rates) are entitled to request more channels to satisfy the minimum data rate threshold. This threshold in conjunction with the RAF has ensured uniform blocking across the areas as well as equal data rates across the user connections. For the parameters chosen it has been shown that the blocking probability in UFCA-II is reduced from 4% to 2.6% when compared with the standard case with no overlap. In Chapter 8, an investigation of the impact of the aerial platform movements in a HAP telecommunication system has been carried out. Results have shown that handoff technique reduces the need for mechanical stabilisation. It has also been shown that handoff mechanism is necessary to ensure continuity of the connections being affected by the platform movements. Furthermore, by employing guard channels it was possible to ensure lower dropping levels but at the expense of higher blocking levels. The immediate handoff scheme based on ABFCA cell overlap model has shown that by allowing overlap, the blocking probability has decreased from 6.5% to 2% compared with the non-overlapping case whilst the dropping probability was also much lower than the no overlap case. It was also shown that the dropping probability has been significantly reduced (when compared with the no overlap case) since the users that required handoff where located in the areas of overlap. Thus, they experienced higher trunking efficiency and as a result the handoff was successful Novel Contributions The original and novel contributions of the work presented in this thesis can be categorised into two major areas. The first is the Channel Assignment related contribution and the second is the Analytical Tool contribution. The major contributions are summarised and presented below. 271

273 CHAPTER 9. Summary and Conclusions Radio Resource Management Related Contributions Random Acceptance Factor Probably the most significant contribution which is used throughout the thesis is the Random Acceptance Factor (RAF). Proving the significance of cell overlap in a communication environment, RAF has been developed to improve the Quality of Service (QoS) both in terms of fairness and blocking probability. The significance of the RAF is that the control mechanism of channel allocation is different from the standard techniques. This is because with the RAF and assuming an unlimited time we have an unlimited granularity. Although we are restricting the number of users coming into the system, we have maintained maximum trunking efficiency. RAF gets round the assignment of discrete integer number of channels to each area or region because RAF is applied over a very large number of conversations. RAF performs the optimisation based on time which is potentially infinite rather than a restricted number of channels. It therefore controls the QoS over time rather than actually restricting channels indefinitely. This work has been presented in [62] and [72]. Exploitation of handoff to reduce the need for mechanical stabilisation mechanism In Chapter 8, it has been shown that handoff can be used as a means to maintain continuity of the communication services when the platform is subjected to a number of movements. From the results it has been shown that by just employing handoff and without any mechanical stabilisation mechanism to compensate for the HAP movements, it is possible to ensure continuity for a large proportion of users. Exploitation of cell overlap to improve overall Blocking and Dropping in a fair way Although L. F. Chang [65] has developed schemes based on overlap within an indoor environment, here we have shown how blocking and dropping can be improved in a fair way by using overlap. More specifically, in Chapter 6 and Chapter 7 it has been shown that when employing cell overlap the blocking probability can be significantly reduced in the areas of overlap. Also by employing the RAF it has been possible to ensure fairness in terms of Quality of Service. In Chapter 8, it has been shown that by employing cell overlap, the number of users that could successfully perform handoff increases as opposed to the case where users connect to the closest virtual base station (i.e. the no overlap case). This was achieved without having to dedicate any channels to the handoff users. Regional Based Channel Allocation In Chapter 6 it has been shown that the channel demand is directly related on the size of the regions formed due to the overlap. It has been shown that performing an optimisation when 272

274 CHAPTER 9. Summary and Conclusions allocating channels to each region it is possible to ensure uniform blocking. It has also been shown that for a small number of channels, the quantisation error is extremely high. As a result the optimisation technique might not be able to match the right number of channels for every region (effect of granularity in the optimisation technique). Limitations of Cell Overlap Cell overlap is primarily interference limited, which means that it cannot be utilised if it expands further than a certain level. From the results presented in Chapter 7 it can be said that for the case the cluster size is 7, the CNIR levels in the overlap regions is adequate to allow users to enter the system. However these users can only be connected with a low bit-rate modulation scheme. If the cell overlap is further expanded, this means that the interference will increase and more users will connect with lower modulation schemes. Therefore, if we wish to maintain a constant bit rate by allowing multiple channel allocation to take place, then it is clear that more users will require more channels to maintain the minimum bit per connection threshold. We have therefore shown that meaningful levels of overlap are available with a HAP communication system. Analytical Tools Three-Dimensional HAP Communication System Simulator The three-dimensional simulator for simulating HAP movements and exploiting cell overlap has been used for the first time to evaluate the performance of a HAP communication system Future Work The work presented in this thesis can be further extended into different areas. These are the following: 1. Dynamic Channel Assignment, Cell Overlap and Random Acceptance Factor It is important that this work extends to other channel allocation schemes such as the Dynamic Channel Allocation (DCA) scheme. This will show the level of improvement of cell overlap can provide to a DCA based scheme and the applicability of the RAF. 2. Minimising Dropping probability arising from handoff failure with Regional based Connection Admission Control (CAC) 273

275 CHAPTER 9. Summary and Conclusions A Connection Admission Control (CAC) policy could be introduced to control the flow of new users on a distance basis. This is to make sure that the traffic supported by the HAP is maintained at the right level so that the dropping probability is minimal when the HAP is moving. In more detail, this technique will allow new users to be admitted if there are less than n-number of users active within a radius y of their (new user) location. If the active users are more than a certain level the new user could be blocked. This is provided that the blocking probability remains within some reasonable limits. 3. Simulation using real time flight data Various mobility models have been examined in this work. Nevertheless from the practical point of view, it would be interesting to try and feed in the code some real time coordinates recorder from a trial. This will produce a set of results based on real movements that can be compared with the ones examined in this thesis. 4. Optimum Cell Size and Optimum Cell Overlap For this work the level of the cell overlap was defined based on the transmit power from the HAP and a minimum received power threshold set on the ground. It is however possible to vary the beamwidth instead of the transmit power to define the extent of the overlap. It would be therefore interesting to find out the level of interference that each of these two techniques performs. Also, what is the optimum cell overlap in which the system performance in terms of the Quality of Service (QoS) is best? It would be also interesting to find out the optimum degree of cell overlap for different cluster sizes. 5. Optimisation of RAF based on more analytical principles In this thesis the RAF has been based on Equation [7.2]. The purpose of the optimisation parameter was to show that it is possible to deliver fairness by using the RAF. Selecting the right parameters would be better if it was done using an analytical technique rather than a numerical one. 274

276 Appendix A. PUBLICATIONS Chapter 10. Bibliography [1] Bell-Laboratories, "High-Capacity Mobile Telephone Systems Technical Report," Submitted to FCC December [2] G. White, Mobile Radio Technology, 1st ed: Butterworth Heinmann, [3] J. D. Gibson, "The Mobile Communications Handbook," IEEE Press, [4] T. S. Rappaport, Wireless communications, principles and practice - Second Edition: Prentice Hall, [5] CCLRC, "The Efficient Use of Broadband Fixed Wireless Access," presented at Broadband Wireless Forum, Cambridge, [6] D. Gesbert, L. Haumonté, H. Bölcskei, R. Krishnamoorthy, and A. J. Paulraj, "Technologies and Performance for Non-Line-of-Sight Broadband Wireless Access Networks," IEEE Communications Magazine, vol. 40, pp , April [7] IEEE, "IEEE Backgrounder," [8] A. Orlowski, "AT&T lifts kimono on WiMAX trials." [9] D. Grace, J. Thornton, T. Konefal, C. Spillard, and T. C. Tozer, "Broadband Communications from High Altitude Platforms - The HeliNet Solution," presented at Wireless Personal Mobile Conference, Aalborg, Denmark, [10] J. Thornton, D. Grace, C. Spillard, T. Konefal, and T. C. Tozer, "Broadband communications from a High-Altitude platform: the European HeliNet programme," IEE, Electronics Communication Engineering Journal, vol. 13, pp , [11] R. Steele, "Guest editorial-an update on personal communications," IEEE Communications Magazine, vol. 30, pp , [12] R. Miura and M. Oodo, "Wireless Communications system using stratospheric platforms," J. Commun. Res. Lab, vol. 48, pp , [13] J.-M. Park, B.-J. Ku, Y.-S. Kim, and D.-S. Ahn, "Technology development for wireless communication system using stratospheric platform in Korea," presented at Proc. IEEE Int. Symp. Personal, Indoor, Mobile Radio Communications, [14] D. Grace, N. E. Daly, T. C. Tozer, A. G. Burr, and D. A. J. Pearce, "Providing Multimedia Communications Services from High Altitude Platforms," International Journal of Satellite Communications 2001, pp , [15] T. C. Tozer and D. Grace, "High-altitude platforms for wireless communications," IEE, Electronics Communication Engineering Journal, vol. 13, pp , [16] T. Konefal, C. Spillard, and G. D., "Site diversity for High Altitude Platforms: A method for the prediction of joint site attenuation statistics," IEE Proceedings on Antennas and Propagation, vol. 149, pp ,

277 Appendix A. PUBLICATIONS [17] C. Spillard, Grace D., J. Thornton, and T. C. Tozer, "Effect of ground station antenna beamwidth on rain scatter interference in High Altitude Platform links," Electron. Lett., vol. 37, pp , [18] ITU, "Recommendation ITU-R F.1500, "Referred characteristics of systems in the fixed service using high altitude platforms operating in the bands GHz and GHz "," vol. ITU-R S.672, [19] M. Oodo, Miura R., T. Hori, T. Morisaki, K. Kashiki, and M. Suzuki, "Sharing and compatibility study between fixed services using High Altitude Platform stations (HAPs) and other services in 31/28 GHz bands," Wireless Personal Communications, vol. 23, pp. 3-14, [20] ITU, "Minimum performance characteristics and operational conditions for High Altitude Platform stations providing IMT-2000 in the Bands MHz, MHz and in the Regions 1 and 3 and MHz and MHz in Region 2. - Reccomendation M.1456," [21] F. Dovis, R. Fantini, M. Mondin, and P. Savi, "Small-scale fading for High Altitude Platform (HAP) propagation channels," IEEE Journal on Selected Areas in Communications, vol. 20, pp , [22] Y. C. Foo, W. L. LIM, and R. Tafazolli, "Performance of High Altitude Platform station (HAPS) in delivery of IMT-2000 W-CDMA," presented at Proc, Stratospheric Platform Systems Workshop, Tokyo, Japan, [23] E. Faletti, M. Mondin, F. Dovis, and G. D., "Integration of a HAP within a terrestrial UMTS network: Interference analysis and cell dimensioning," Wireless Personal Communications - Special Issue on Broadband Mobile Terrestrial-Satellite Integrated Systems, vol. 24, pp , [24] S. Masumura and M. Nakagawa, "Joint system of terrestrial and High Altitude Platform stations (HAPS) cellular for W-CDMA mobile communications," IEICE Trans. Commun, vol. E85-B, pp , [25] I. Katzela and M. Naghshineh, "Channel Assignment Schemes for Cellular Mobile Telecommunication Systems: A comprehensive Survey," IEEE Personal Communications, pp , [26] G. M. Djuknic, J. Freidenfelds, and Y. Okunev, "Establishing wireless communications services via high-altitude aeronautical platforms: a concept whose time has come?" in IEEE Communications Magazine, vol. 35, [27] J.-P. Linnartz, Narrowband Land-Mobile Radio Networks. London: Artech House, [28] CAPANINA, "CAPANINA - Framework 6 European Union Funded Project," [29] C. Spillard, J. D. Penin, M. Mondinm, and E. Faletti, "Topology and Mobility Effects on Links," University of York, York FP6-IST , 28 Sep [30] M. Sreetharan and R. Kumar, "Cellular Digital Packet Data," pp ,

278 Appendix A. PUBLICATIONS [31] V. H. MacDonald, "The Cellular Concept," Bell System Technical Journal, vol. 58, pp , [32] T. S. Rappaport, Wireless communications, principles and practice: Prentice Hall, [33] W. C. Y. Lee, Mobile Cellular Telecommunications Systems, vol New York: McGraw-Hill, [34] L. Ortigoza-Guerrero and D. Lara-Rodriguez, "CPMCB: A Suitable DCA Scheme for the Pan-European GSM System," presented at Proc. IEEE Veh. Tech. Conf. VTC'97, Phoenix, Arizona, US, [35] M. Zhang and T.-S. P. Yum, "Comparison of Channel Assignment Strategies in Cellular Mobile Telephone Systems," IEEE Transactions on Vehicular Technology, vol. 38, pp , [36] J. C.-I. Chuang, "Performance Issues and Algorithms for Dynamic Channel Assignment," Ieee Journal on Selected Areas in Communications, vol. 11, pp , [37] M. M.-L. Cheng and J. C.-I. Chuang, "Performance Evaluation of Distributed Measurement-Based Dynamic Channel Assignment in Local Wireless Communications," IEEE Journal on Selected Areas in Communications, vol. 14, [38] D. Akerberg, "On Channel Definitions and Rules for Continuous Dynamic Channel Selection in Coexistence Etiquettes for Radio Systems," presented at 44th IEEE Vehicular Technology Conference, Stockholm, [39] B. Eklundh, "Channel Utilization and Blocking Probability in a Cellular Mobile Telephone System with Directed Retry," IEEE Transactions on Communications, vol. COM-34, pp , [40] D. Everitt, "Traffic Capacity of Cellular Mobile Communications Systems," Comp. Networks ISDN Sys., vol. 20, pp , [41] X. Lagrange and B. Jabbari, "Fairness in wireless microcellular networks," Vehicular Technology, IEEE Transactions on, vol. 47, pp , [42] D. Grace, G. Chen, G. White, J. Thornton, and T. C. Tozer, "Improving the System Capacity of mm-wave Broadband Services Using Multiple High Altitude Platforms," presented at IEEE Global Communications Conference (GLOBECOM), San Francisco, USA, [43] D. Grace, C. Spillard, and T. C. Tozer, "High Altitude Platform Resource Management Strategies with Improved Connection Admission Control," presented at IEEE Wireless Personal Multimedia Communications Conference (WPMC), Yokosuka, Japan, [44] J. Thornton, "A Low Sidelobe Asymmetric Beam Antenna for High Altitude Platform Communications," IEEE Microwave and Wireless Components Letters, vol. 14, [45] G. White, A. G. Burr, and T. Javornik, "Modeling of nonlinear distortion in broadband fixed wireless access systems," IEE Electronics Letter, vol. 39, pp , [46] D. Gruber, "Do We Really Need Quaternions?" vol. 2002: Ted Gruber Software,

279 Appendix A. PUBLICATIONS [47] A. S. Glassner, Graphics Gems: Academic Press, [48] J. Thornton and D. Grace, "Effect of Lateral Displacement of a High Altitude Platform on Cellular Interference and Handover," IEEE Transactions on Wireless Communications, [49] L. Bostock, S. Chandler, and C. Rourke, Further Pure Mathematics: Stanley Thornes (Publishers) Ltd., [50] J. Thornton, D. Grace, M. H. Capstick, and T. C. Tozer, "Optimising an Array of Antennas for Cellular Coverage from a High Altitude Platform," IEEE Transactions on Wireless Communications, vol. 2, pp , [51] N. E. Daly, D. Grace, T. C. Tozer, D. A. J. Pearce, and A. G. Burr, "Prediction of Frequency Reuse Behaviour for High Altitude Platforms," presented at Airship Convention 2000, [52] J. Thornton, D. A. J. Pearce, D. Grace, M. Oodo, K. Katzis, and T. C. Tozer, "Effect of Antenna Beam Pattern and Layout on Cellular Performance in High Altitude Platform Communications," International Journal of Wireless Personal Communications, [53] D. A. J. Pearce, "Improving Spectrum Efficiency in Fixed Cellular Communication Systems," in Department of Electronics. York: University of York, [54] S. S. Solutions, "Select Yourdon," 3.10 Educational ed, [55] R. S. Pressman, Software Engineering: A Practitioner's Approach - European 3rd Rev.: McGraw-Hill Book Company, [56] D. Grace and J. Thornton, "System Level Design Aspects for the Delivery of Broadband Services over Helinet," University of York, York T1 Report, [57] W. H. Press, S. A. Teukolsky, W. T. Vetterling, and B. P. Flannery, Numerical Recies in C: Cambridge University Press, [58] L. Kleinrock, Queueing Systems Volume I: Theory, [59] D. Grace, C. Spillard, J. Thornton, and T. C. Tozer, "Channel Assignment Strategies for a High Altitude Platform Spot-Beam Architecture," presented at IEEE Personal, Indoor, and Mobile Radio Communications Conference (PIMRC), Lisbon, Portugal, [60] C. A. Balanis, Antenna Theory Analysis and Design. New York: John Wiley & Sons, Inc, [61] E. W. Weisstein, "Conic Section," vol. 2003: MathWorld-A Wolfram Web Resource, [62] K. Katzis, D. Grace, and D. A. J. Pearce, "Fixed Channel Allocation Techniques Exploiting Cell Overlap for High Altitude Platforms," presented at European Wireless Conference, Barcelona, Spain, [63] M. R. Spiegel, Statistics, Chapter: Statistical Estimation Theory, [64] I. Sommerville, Software Engineering, 6th Edition ed,

280 Appendix A. PUBLICATIONS [65] L. F. Chang, A. R. Noerpel, and A. Ranade, "Performance of Personal Access Communications System - Unlicensed B," IEEE Journal on Selected Areas in Communications, vol. 14, pp , [66] D. Grace, G. Chen, G. White, J. Thornton, and T. C. Tozer, "Improving System Capacity of Broadband Services Using Multiple High Altitude Platofmrs," IEEE Transactions on Wireless Communications, vol. 4, pp , [67] T. Fujii, "Selective Handover for Traffic Balance in Mobile Communications," presented at IEEE / Supercomm / ICC92, [68] Y. Liu, D. Grace, and P. D. Mitchell, "Effective System Spectral Efficiency Applied to a Multiple High Altitude Platform System," IEE Proceedings on Communications, [69] C. E. Shannon, "A Mathematical Theory of Communication," The Bell System Technical Journal, vol. 7, pp , [70] C. Bettstetter, "Smooth is Better than Sharp: A Random Mobility Model for Simulation of Wireless Networks.," presented at Proceedings of the 4th ACM International Workshop on Modeling, Analysis, and Simulation of Mobile Systems (MSWIM), [71] D. Grace, "Distributed Dynamic Channel Assignment for the Wireless Environment," in Department of Electronics, vol. D.Phil. York: University of York, 1998, pp [72] K. Katzis, D. Grace, and D. Pearce, "Fairness in Channel Allocation in a High Altitude Platform Communication System exploiting Cellular Overlap," presented at Wireless Personal Multimedia Communication Conference (WPMC), Abano Terme, Italy, [73] L. J. Ehernberger, "Stratospheric Turbulence Measurements and Models for Aerospace Plane Design," NASA, California [74] M. H. Capstick and D. Grace, "High Altitude Platform mm-wave Aperture Antenna Steering Solutions," International Journal of Wireless Personal Communications, Accepted [75] D. I. Axiotis, M. E. Theologou, and E. D. Sykas, "The Effect of Platform Instability on the System Level Performance on HAPS UMTS," IEEE Communications Letters, vol. 8, [76] S. Kalyanasundaram, E. K. P. Chong, and N. B. Shroff, "An Efficient Scheme to Reduce Handoff Dropping in Leo Satellite Systems," Wireless Networks, vol. 7(1), pp , [77] J. Tajima and K. Imamura, "A strategy for flexible channel assignment in mobile communication systems," IEEE Trans on Vehicular Technology, vol. 37, pp , [78] D. C. Cox and D. O. Reundnik, "Increasing Channel Occupancy in Large-Scale Mobile Radio Systems: Dynamic Channel Assignments," IEEE Trans on Vehicular Technology, vol. 22, [79] R. Guerin, "Queueing Blocking System with Two Arrival Streams and Guard Channels Guard Channels," IEEE Transactions on Communications, vol. 36, pp ,

281 Appendix A. PUBLICATIONS [80] J. Zander, S.-L. Kim, M. Almgren, and O. Queseth, Radio Resource Management for Wireless Networks: Artech House Books, [81] C. G. Cassandras, L. Dai, and C. G. Panayiotou, "Ordinal Optimisation for a Class of Deterministic and Stochastic Discrete Resource Allocation Problems," IEEE TRansactions on Automatic Control, vol. 43, [82] J. Karlsson and B. Eklundh, "A Cellular Mobile Telephone System with Load sharing - An Enhancement of Directed Retry," IEEE Transactions on Communications, vol. 37, pp. May 1989, [83] C. Posner and R. Guerin, "Traffic Policies in Cellular Radio that Minimize Blocking of Handoffs," ITC-II, [84] R. Singh, S. M. Elnoubi, and C. Gupta, "A New Frequency Channel Assignment Algorithm in High Capacity Mobile Communications Systems," IEEE Trans on Vehicular Technology, vol. 31, [85] S. Tekinay, "A Measurement-Based Prioritization Scheme for Handovers in Mobile Cellular Networks," IEEE JSAC, vol. 1, pp , [86] D. Grace, K. Katzis, D. A. J. Pearce, and P. D. Mitchell, "Rapid MAC-Layer Handoff for Broadband Communications from High Altitude Platforms (to be submitted)," in IEEE Transactions on Communications: University of York, [87] D. I. Axiotis and M. E. Theologou, "Modeling the positional instabilities of High Altitude Stratospheric Platform stations," AIAA J. Aerospace Computing, Inform., Commun.,

282 Appendix A. PUBLICATIONS Appendix A - Publications A.1 Full list of publications and invited talk presentations 282 Selected Publications A.2 5 th European Wireless Conference Mobile and Wireless Systems beyond 3G 283 (EW2005) A.3 Wireless Personal Multimedia Communications Conference (WPMC) 290 A.4 International Workshop on High Altitude Platform Systems (WHAPS)

283 Appendix A. PUBLICATIONS A.1. Full list of publications and invited talk presentations Conference Publications [1] K. Katzis, D. J. Pearce, D. Grace, Comparison between resource allocation techniques for high altitude platforms and ground based cellular systems, PREP 2002, April 2002, Nottingham, UK [2] K. Katzis, D. J. Pearce, D. Grace, Channel Allocation Techniques for high altitude platforms, PREP 2003, April 2003, Exeter, UK [3] K. Katzis, D. J. Pearce, and D. Grace, Fixed Channel Allocation Techniques Exploiting Cell Overlap for High Altitude Platforms, The Fifth European Wireless Conference Mobile and Wireless Systems beyond 3G, Barcelona, Spain, February [4] K. Katzis, D. A. J. Pearce, and D. Grace, Fairness in Channel Allocation in a High Altitude Platform Communication System exploiting Cellular Overlap, accepted for presentation at the Wireless Personal Multimedia Communications Conference (WPMC), Abano Terme, Italy, September 2004 [5] K. Katzis, D. Grace, and D. A. J. Pearce, Impact of High Altitude Platform movements on Cellular Handover, accepted for presentation at the International Workshop on High Altitude Platform Systems (WHAPS), Athens, Greece, September Journal Publications [1] J Thornton, D. A. J. Pearce, D. Grace, M. Oodo, K. Katzis,T. C. Tozer, Effect of Antenna Beam Pattern and Layout on Cellular Performance in High Altitude Platform Communications, International Journal of Wireless Personal Communications, accepted March Invited Talks [1] An overview of High Altitude Platforms and Terrestrial communication systems while focusing on resource allocation techniques, King s College London (July 2002) [2] High Altitude Platforms & Resource Allocation Techniques, University of Cyprus (March 2004). 282

284 Appendix A. PUBLICATIONS A.2. 5 th European Wireless Conference Mobile and Wireless Systems beyond 3G (EW2005) 283

285 Appendix A. PUBLICATIONS 284

286 Appendix A. PUBLICATIONS 285

287 Appendix A. PUBLICATIONS 286

288 Appendix A. PUBLICATIONS 287

289 Appendix A. PUBLICATIONS 288

290 Appendix A. PUBLICATIONS 289

291 Appendix A. PUBLICATIONS A.3. Wireless Personal Multimedia Communications Conference (WPMC) 290

292 Appendix A. PUBLICATIONS 291

293 Appendix A. PUBLICATIONS 292

294 Appendix A. PUBLICATIONS 293

295 Appendix A. PUBLICATIONS 294

296 Appendix A. PUBLICATIONS A.4. International Workshop on High Altitude Platform Systems (WHAPS) 295

297 Appendix A. PUBLICATIONS 296

298 Appendix A. PUBLICATIONS 297

299 Appendix A. PUBLICATIONS 298

300 Appendix A. PUBLICATIONS 299