ABSTRACT
The aim of this study is to develop and analytically compare Quotients Regression based empirical models with artificial neural network (ANN) based models for path loss prediction. The terrains considered as case study include i) the rural terrain between Jos and Abuja ii) the Urban terrain (Abuja), and iii) the semi-urban terrain (Maiduguri). The empirical models considered include the Okumura, Hata-Okumura, COST 231 Hata and the COST 231Walfisch-Ikegami, while the two types of ANN include the Multilayer Perceptron Neural Network (MLP-NN) and the Generalised Radial Basis Function Neural Network (GRBF-NN). A Quotients Regression Technique (QRT) for empirical model adaptation was developed and used to selectively adapt these empirical models to the terrains, based on path loss measurements obtained from Base Transceiver Stations (BTS) situated within the terrains. The adaptation accuracy of the QRT was determined through comparisons with two existing adaptation techniques: i) The Okumura GAREA (Gain due to type of environment) correction factor, and ii) The Root Mean Squared Error (RMSE) Adaptation Technique (RAT). The comparative analysis of the QRT adapted empirical models with ANN-based models was based on three distinct approaches: i) Splitting path loss data into 60% Training, 10% Validation and 30% Testing, ii) Splitting path loss data into 50% Training Set and 50% Testing Set. iii) Random training with path loss data from one BTS and testing with data from another. The following results were obtained: i) The QRT has the highest adaptation accuracy, based on combined RMSE (Root Mean Squared Error) value of 2.1dB across the three terrains, the RAT technique has 5.66dB across the three terrains, while the Okumura GAREA has 8.95dB across the rural terrain alone, ii) The ANN-based models have the highest path loss prediction accuracy, based on an RMSE value of 3.98dB, followed by the QRT adapted empirical models with 4.49dB. The RAT adapted empirical models have 5.83dB, while the empirical models are the least accurate with 7.07dB. By implication, the ANN-based models only slightly outperform the QRT adapted empirical models by 0.51dB. However, the QRT adapted empirical models have the highest R2 (coefficient of determination) value of 0.81, and by implication, the best fit resulting from the best correlation with the measured path loss data. On the other hand, the QRT adapted empirical models offer an improvement of about 1.34dB over the RAT adapted empirical models, as well as an improvement of 2.58dB over existing Empirical Models. The proximity in performance of the QRT adapted empirical models to the ANN-based models can be attributed to the efficiency of the QRT.
1
TABLE OF CONTENTS
TITLE PAGE i DECLARATION ii CERTIFICATION iii DEDICATION iv ACKNOWLEDGEMENTS v TABLE OF CONTENTS vi LIST OF FIGURES xiii LIST OF TABLES xv LIST OF ABBREVIATIONS AND SYMBOLS xvii ABSTRACT xxii CHAPTER ONE: INTRODUCTION 1.1 Background 1 1.1.1 The Cellular Network Concept 3 1.1.2 GSM Network Architecture 5 1.1.3 Evolution of Cellular Technology 6 1.2 Aim and Objectives 11 1.3 Statement of Problem 12 1.4 Methodology 14 1.5 Significance of the Study 15 1.6 Thesis Outline 15
ix
CHAPTER TWO: LITERATURE REVIEW 2.1 Introduction 16 2.2 Review of Fundamental Concepts 17 2.2.1 Radio Propagation Mechanisms 17 2.2.1.1 Free Space Attenuation 17
2.2.1.2 Diffraction 19
2.2.1.3 Scattering 23
2.2.1.4 Reflection 23
2.2.1.5 Transmission 24
2.2.1.6 Refraction 24
2.2.1.7 Multipath Propagation and Fading 25 2.2.1.8 Absorption 29 2.2.2 Radio Propagation Models 30 2.2.2.1 Deterministic Models 30 2.2.2.2 Stochastic Models 31 2.2.2.3 Empirical Models 31 2.2.3 Soft Computing 36 2.2.3.1 Artificial Neural Networks 38 2.2.3.2 The Fundamental Unit of Neural Networks 39 2.2.3.3 Activation Functions 40 2.2.3.4 Learning 42
2.2.3.5 Training Algorithms 44
2.2.3.6 Artificial Neural Network Architectures 45
x
2.2.3.7 Artificial Neural Network Parameters 52 2.2.4 Discrete Least Squares Approximation 53 2.2.5 Types of Prediction Area 55 2.2.6 Performance Evaluation Statistics 55 2.2.6.1 Absolute Mean Error 55 2.2.6.2 Standard Deviation 56 2.2.6.3 Root Mean Squared Error 56 2.2.6.4 Coefficient of Determination (Goodness of Fit) 57 2.3 Review of Similar Works 57 CHAPTER THREE: MATERIALS AND METHODS 3.1 Introduction 67 3.2 Received Power Measurement and Path Loss Computation 67 3.3 Determination of reliabilities of empirical models for path loss Prediction 69 3.4 Development of the Proposed Quotients Regression Technique 70 3.5 Adaptation Accuracy Comparison of the Quotients Regression Technique with Existing Adaptation Techniques 72 3.5.1 The Okumura Adaption Technique using GAREA 72 3.5.2 The Root Mean Squared Error Adaptation Technique (RAT) 73 3.6 Creating the Artificial Neural Network Predictors 73 3.6.1 Creating the MLP-NN Based Model 74
3.6.2 Creating the GRBN-NN Based Model 75 3.7 Comparison of Quotients Regression Adapted Empirical Models with Artificial Neural Network Predictors 76
xi
CHAPTER FOUR: RESULTS AND ANALYSES 4.1 Introduction 78 4.2 Comparison of Adaptation Techniques Using the Okumura Model 80 4.2.1 Adapting the Okumura Model using Curve Correction Factors 80 4.2.2 Adapting the Okumura Model using the RMSE Adaptation Technique (RAT) 81 4.2.3 Adapting the Okumura Model using the Quotients Regression Technique (QRT) 81 4.2.4 Comparison of Adaption Techniques as applied to the Okumura Model 83 4.2.5 Generalization Test for Adapted Okumura Models 84 4.3 Comparison of Quotients Regression Adapted Empirical Models with Artificial Neural Network Predictors 86 4.3.1 The Rural Area between Jos and Abuja 86 4.3.1.1 Testing the COST 231 Hata and the Hata-Okumura models for acceptability 87 4.3.1.2 Adapting the COST 231 Hata and the Hata-Okumura Models 88 4.3.1.3 Comparison of Adaptation Techniques 90 4.3.1.4 Generalisation test for RAT Adapted and QRT Adapted Empirical Models 92 4.3.1.5 Comparison of ANN-based and Quotients Regression Adapted Empirical Models using the Training-Validation-Testing Technique . . . 94 4.3.1.6 Comparison of ANN-based Models with QRT Adapted Empirical Models using the 50%Training and50% Testing Technique . . . 96 4.3.1.7 Comparison of Model Predictors by Random Training with one Base Station and Testing with another . . . . . . 99 4.3.1.8 Combined performance Analysis based on three Comparative Techniques 101 4.3.2 The Urban Terrain (Abuja) 102 4.3.2.1 Testing the COST 231 Walfisch-Ikegami and the COST 231 Hata models for acceptability . . . . . . . 103
4.3.2.2 Adapting the COST 231 Walfisch-Ikegami and the COST 231 Hata Models 104
xii
4.3.2.3 Comparison of Adaptation Techniques 105 4.3.2.4 Performance Comparison of RAT Adapted and QRT Adapted Empirical Models . . . . . . . . . 105 4.3.2.5 Comparison of ANN-based models with the QRT Adapted Empirical Models using the Training-Validation-Testing Technique . . . . 107 4.3.2.6 Comparison of ANN-based models with the QRT Adapted Empirical Models using the 50% Training and 50% Testing Technique 108 4.3.2.7 Comparison of ANN-based models with the QRT Adapted Empirical Models by Random Training with one Base Station and Testing with another . . . 110 4.3.2.8 Combined Performance Analysis across the three Techniques 112 4.3.3 The Semi-Urban Terrain (Maiduguri) 113 4.3.3.1 Acceptability Test for the COST 231 Walfisch-Ikegami and COST 231 Hata 114 4.3.3.2 Adapting the COST 231 Walfisch-Ikegami and the COST 231 Hata Models 115 4.3.3.3 Comparison of Adaptation Techniques . . . . . 116 4.3.3.4 Performance Comparison of RAT Adapted and QTR Empirical Models 117 4.3.3.5 Comparison of ANN-based models with the QRT Adapted Empirical Models using the Training-Validation-Testing Technique . . . . 118 4.3.3.6 Comparison of ANN-based models with the QRT Adapted Empirical Models using the 50% Training and 50% Testing Technique . . . 120 4.3.3.7 Comparison of ANN-based models with the QRT Adapted Empirical Models by Random Training with one Base Station and Testing with another . 122 4.3.3.8 Combined performance Analysis across three Techniques 124 4.4 Overall Performance Comparison of the Adaptation Techniques across the three Terrains . . . . . . . . . 125 4.5 Overall Performance Comparison of Model Categories across the three Terrains . . . . . . . . . 126
xiii
CHAPTER FIVE: SUMMARY, RECOMMENDATION AND CONCLUSION 5.1 Introduction 128 5.2 Summary 128 5.3 Significant Contributions 129 5.4 Conclusions 130 5.5 Recommendations further work 131 5.6 Limitations 132 REFERENCES 133 APPENDICES 145 APPENDIX A: Measured Received Power and Computed Path Loss/Mobile Network Parameters . . . . . . . 145 Table I: Rural Area between Jos and Abuja 145 Table II: Urban Terrain (Abuja) 146 Table III: Semi-Urban Terrain (Maiduguri) 147 APPENDIX B: MATLAB CODE APPENDIX B1: Comparison of Adaptation Techniques 148 APPENDIX B2: Applicability Test and Adaptation of Empirical Models (Rural) . . . . . . 151 APPENDIX B3: Technique A (Rural Terrain) 156 APPENDIX B4: Techniques B and C (Rural) 160 APPENDIX B5: Applicability Test and Adaptation of Empirical Models (Abuja) . . . . . . . 165 APPENDIX B6: Technique A (Abuja) 171 APPENDIX B7: Techniques B and C (Abuja) 176
CHAPTER ONE
INTRODUCTION 1.1 Background Since the introduction of cellular systems for mobile communications decades ago, they have undergone tremendous growth and evolution. Over the decades the number of mobile subscribers around the globe has also undergone tremendous growth and will continue to grow. Hence, the availability of high capacity mobile networks with quality delivery of service has become very important especially in recent times. As a result, the creation of a mobile network with good network coverage demands proper planning. One of the greatest challenges faced by radio systems engineers is the accurate determination of the radio propagation characteristics of a particular terrain, with a view to designing a well engineered radio path. When planning a cellular communications network, it is necessary to define the optimum number of base stations within a coverage area and to also resolve other network related problems. Hence, it is necessary to determine the characteristics of radio signals within the limits of the service area. Since various terrains create specific conditions for propagation of the radio waves within a service area, detailed knowledge of channels propagation characteristics of radio signals is a necessary precondition for the development of effective communication systems operating in any environment. This stems from the fact that as radio signals propagate from the base station to the mobile station, some power is lost, and this power loss is known as path loss. Path loss is dependent on the carrier frequency, antenna height and distance between mobile station and base station (Ashis, 2012). There is a variety of models used by researchers and radio engineers to predict path loss across a particular terrain. A model that suits a given terrain (or environment) may not necessarily be suitable to another terrain. Hence, radio
2
engineers and researchers have been evolving new techniques for accurate prediction of path loss. Researchers for many years have made attempts to either develop new propagation models or modify existing ones. Empirical models are some of the most widely used radio propagation models. However, it is pertinent to note that these empirical models are simple to implement but not universally applicable due to terrain clutter differences across the globe. As such, researchers over the years have developed various techniques for adapting empirical models to suite terrains. The most popular technique used by researchers in recent times have had to do with the use of computed correction factors to achieve adaptation as demonstrated by Nadir et al (2010), Ubom et al (2011), Ogbulezie et al (2013), etc. However, in certain scenarios, this technique may not provide the best possible fit to the empirical model if the slope component of the empirical model significantly differs from that of the function of the best fit curve through measured path loss points. Hence, it necessary to create a technique that does not only guarantee the best possible fit for the empirical model, but also ensures the adapted model is robust when tested using new data.
Structural impairments along radio propagation paths demand that non-linear approximation functions are used for path loss prediction. Empirical models are linear approximation based functions; hence they not very accurate in predicting path loss along terrains with well diversified structural impairments, demanding the implementation of non-linear functions that can lead to greater prediction accuracy. As described in (Faria et al., 2009), neural networks can learn to approximate any function to a given accuracy and behave like associative memories by using just example data that is representative of the desired task, operating then as model free estimators. This gives them a key advantage over traditional
3
approaches to function estimation such as the statistical methods. Hence, computational intelligence techniques have been applied recently to predict path loss with greater accuracy as demonstrated by Ostlin (2010), Ignacio et al. (2012), Abraham et al. (2014), Joseph et al. (2014), Callistus et al. (2015) etc. 1.1.1 The Cellular Network Concept The cellular network has become the most common platform for wireless communications. With a mobile phone, one can make calls and access data from almost any location across the globe. A cellular network is a communications system whose service area is divided into operating zones called cells, inside which communication between mobile and base stations is carried out by a radio channel. Switching equipment is used to interconnect different parts of a mobile network and to also allow access to the fixed Public Switched Telephone Network (PSTN) (Tarun et al, 2013). The introduction of the cellular concept played a great role in resolving the spectral congestion and user‟s capacity problems (Ashis, 2012). The cellular concept splits a given service area into cells, each served by a Base Station (BS), thereby enabling the frequency reuse concept.
The high network capacity of modern cellular networks stems from the frequency reuse concept as frequency spectrum allocation to cellular networks is very limited (Rakibul et al., 2011). As such, the coverage area is divided into cells, each of which is served by a BS. Each BS (or cell) is assigned a group of frequency bands or channels. These are assigned to subscribers on demand. In or to avoid radio co-channel interference, the group of channels assigned to neighbouring cells must be different. However, distant cells with insignificant co-channel interference between them can be assigned the same group of channels. Typically, seven
4
neighbouring cells are grouped together to form a cluster. Cells are group in sevens to form a cluster, and hence and the total available channels are divided into seven groups, each of which is assigned to a cell. As shown in Figure 1.1, cells marked with the same identifier have the same group of channels assigned to them. Furthermore, the cells marked with different identifiers must be assigned different groups of channels.
Figure 1.1: A Cellular Network Structure showing Frequency Reuse (Lei et al., 2004) Jingyuan and Ivan (2005) further explain that during communication a Mobile Station (MS) registers with the nearest BS while the corresponding Mobile Services Switching Center (MSC) stores the information about that MS and its position. This information is used to direct incoming calls to the MS. The concept behind mobile communication is based on the hand-over process. Hand-over occurs during mobile communication when the MS moves from its serving cell to a neighboring cell, thereby forcing a change of frequency since neighboring cells use different channels. The BS constantly monitors any decrease in signal power as the MS approaches the edge of a cell, and compares the signal strength with the signals coming from adjacent cells, after which the call is handed over to the cell with the strongest signal. During the switch, the line is lost for about 400ms. When the MS moves to a different location, it registers itself at the new MSC, and the information on the MS is constantly is updated, such that the MS can be used outside of its original location.
5
1.1.2 GSM Network Architecture The GSM (Global System for Mobile Communications) is a typical cellular network. Introduced in the 1990’s the GSM is currently the most popular mobile phone across the globe especially the third world countries. The GSM network architecture shown in figure 1.2, as described by Sneha et al. (2014) includes the following components:
Figure 1.2: GSM (3G) Architecture (Sneha et al., 2014)
i) The Mobile Station (MS) is basically a mobile phone that contains the Subscriber Identity Module (SIM), which contains relevant user information.
ii) The Base Station Subsystem (BSS) provides the interface between the MS and the NSS. It handles transmission and reception. The BSS comprises of the following:
1. Base Transceiver Station (BTS) or Base Station: the BTS is found in the centre of a cell and it is mapped to transceivers and antennas used in given cell within the network. Its transmitting power defines the size of a cell.
6
2. The Base Station Controller (BSC): It is responsible for controlling a group of BTSs and also managing their radio resources. A BSC handles handoffs, frequency hopping, exchange functions and power control over each managed BTSs.
iii) The Network and Switching Subsystem (NSS): The NSS manages communications between a set of mobile users and other mobile users, Integrated Systems Digital Network (ISDN) users, fixed telephony users, etc. It also contains a database for storing information about subscribers and mobility management. Details on the components that make up the NSS can be obtained from Sneha et al., (2014).
The GSM network as shown in figure 1.2 comprises three interfaces: the Um, the A-bis and the A. The Um is the Radio interface between MS and BTS. The A-bis is the interface between BTS and BSC. Its primary functions include traffic channel transmission, terrestrial channel management, and radio channel management. The interface between the BSS and the NSS is called the A interface. 1.1.3 Evolution of Cellular Technology With technological advancement, the cellular technology has evolved through generations, namely 1G, 2G, 2.5G, 3G and 4G. The fifth generation (5G) is the next wireless communication standard.
a) First Generation (1G)
The First Generation (1G) of cellular technology was essentially an analogue wireless telecommunication system used for voice calls using cell phones (Mudit and Anand, (2010). The
7
1G standards used included the NMT (Nordic Mobile Telephone), used in Nordic countries, Eastern Europe and Russia. Others include AMPS (Advanced Mobile Phone System) used in the United States, TACS (Total Access Communications System) in the United Kingdom, C-Netz in West Germany, Radiocom 2000 in France, and RTMI in Italy.
b) Second Generation (2G)
As described in Mudit and Anand, (2010), the second generation (2G) was digital and made use of the Time Division Multiple Access (TDMA) and Code Division Multiple Access (CDMA) concepts to increase the network capacity. The 2G network had improved security features and accommodated voice coding and encryption. Popular 2G standards included the GPRS (General Packet Radio Service) classified as 2.5G and EDGE (Enhanced Data rates for GSM Evolution) known as 2.75G. EDGE is an upgrade over GPRS and can function on any network with GPRS deployed on it, provided the carrier implements the necessary upgrades. EDGE technology carries packet switch data and circuit switch data and at a faster rate than GSM. GPRS could provide data rates ranging from 56 kbit/s to 115 kbit/s. 2G provided services such as Wireless Application Protocol (WAP) access, Multimedia Messaging Service (MMS), and for Internet communication services such as email and World Wide Web access.
c) Second and Half Generation (2.5G)
The Second and Half Generation as the name implies came into existence between the second generation (2G) and third generation (3G). As described in Mudit and Anand (2010), the term 2.5G refers to 2G-systems that combine a packet switched domain a circuit switched domain. The term “2.5G” is an informal term, invented solely for marketing purposes, unlike “2G” or
8
“3G” which are officially defined standards based on those defined by the International Telecommunication Union (ITU). The standard used on 2.5G was the GPRS, which provided data rates from 56 kbit/s to 115 kbit/s. It provided services such as Wireless Application Protocol (WAP) access, Multimedia Messaging Service (MMS), and for Internet communication services such as email and World Wide Web access. 2.5G networks also provided services such as WAP, MMS, SMS mobile games, and search and directory.
d) Third Generation (3G)
As described in Mudit and Anand (2010), the third generation (3G) was aimed at providing high-speed packet-switching data transmission in addition to circuit-switching voice transmission across the Internet. Common 3G network services include wireless voice telephony, video calls/ teleconferencing, broadband wireless data, GPS (global positioning system), mobile television etc. The standard implemented is the High-Speed Packet Access (HSPA) capable of providing data transmissions speeds up to 14.4Mbit/s on the downlink and 5.8Mbit/s on the uplink. 3G technologies make use of TDMA and CDMA. 3G Technology was developed for fast data transfer rates. High-Speed Downlink Packet Access (HSDPA) is a 3.5G is a packet-based Wireless CDMA (W-CDMA) downlink data service with data transmission up to 8-10 Mbit/s and 20 Mbit/s for Multiple-Input Multiple-Output (MIMO) systems over at 5MHz. HSDPA is implemented as Adaptive Modulation and Coding (AMC), MIMO, Hybrid Automatic Request (HARQ), fast cell search, and advanced receiver design. Another standard is the 3.75G High-Speed Uplink Packet Access (HSUPA) capable of higher transfer rates.
9
e) Fourth Generation 4G
4G refers to the fourth generation of cellular wireless standards. It is basically the extension of the 3G technology with more bandwidth and services and offers high quality audio/video streaming over end to end Internet Protocol (Mudit and Anand, 2010). According to Shin et al. (2013), the fourth generation long-term evolution Advanced (4G LTE Advanced) is the latest standard in the development of 4G mobile networks and has been designed to offer users of 4G mobile devices much faster data speeds than those on offer from existing 4G LTE networks, making it a true 4G standard. According to Arun et al., (2013), 4G networks are essentially based on packet switching technology. The fourth generation mobile systems use orthogonal frequency division multiplexing (OFDM), MIMO, software defined radio (SDR) technologies (Patil et al, 2012). The modulation techniques implemented enhance efficiency by dividing 5, 10 or 20 MHz wide channels into smaller sub channels or subcarriers each 15 kHz wide. Each is modulated with part of the data. The modulation techniques used are QPSK (Quadrature Phase Shift Keying) or 16QAM (Quadrature amplitude modulation). With the aid of MIMO operation that uses several transmitter-receiver-antennas, the data stream is divided amongst the antennas to boost speed and to make the link more reliable. With the help of OFDM and MIMO, LTE can deliver data rates up to 100 Mb/s downstream and 50 Mb/s upstream under the best conditions. In 4G the theoretical upper data rate is 1Gb/s.
In contrast to the GSM network architecture, the LTE- Advanced Architecture comprises of essentially two layers: the access network known as the Evolved Universal Terrestrial Radio Access Network (E-UTRAN) and the core network known as the Evolved Packet Core (EPC) network as shown in Figure 1.3 (Alcatel, 2013). As described in (Alcatel, 2013), the term “LTE”
10
encompasses the evolution of the Universal Mobile Telecommunications System (UMTS) radio access through the E-UTRAN, and it is accompanied by an evolution of the non-radio aspects under the term “System Architecture Evolution” (SAE), which includes the EPC network.
Figure 1.3: LTE-Advanced E-UTRAN network architecture (Ghassan et al., 2012) Together LTE and SAE comprise the Evolved Packet System (EPS). EPS uses the concept of EPS bearers to route IP traffic from a gateway in the PDN to the UE (User Equipment). A bearer is an IP packet flow with a defined quality of service (QoS) between the gateway and the UE. The E-UTRAN and EPC together set up and release bearers as required by applications. Details on the LTE-Advanced Architecture can be obtained from Alcatel (2013).
f) Fifth Generation (5G)
As described in Ekram et al., (2014), the evolving fifth generation (5G) cellular wireless networks are targeted towards providing higher data rates, excellent end-to-end performance and better user-coverage in hot-spots and crowded areas with lower latency, lower energy
11
consumption and cost per information transfer, compared with existing cellular networks. In order to address these challenges, 5G systems will be based on a multi-tier architecture consisting of macrocells, different types of licensed small cells, relays, and device-to-device (D2D) networks to serve users with different quality-of-service (QoS) requirements in a spectrum and energy-efficient manner. The features of 5G will include wireless networks network tiers of different sizes, transmit powers, backhaul connections, different radio access technologies that are accessed by an unprecedented number of smart and heterogeneous wireless devices. The 5G architectural upgrade along with the advanced physical communications technology such as high-order spatial multiplexing multiple-input multiple-output (MIMO) communications will provide higher network capacity for more simultaneous users, or higher level spectral efficiency, when compared to the 4G networks. Radio resource and interference management will be a key research challenge in multi-tier and heterogeneous 5G cellular networks. Details on the prospects and challenges of 5G cellular networks are presented in Ekram et al., (2014). 1.2 Aim and Objectives The aim of this study is to develop the Quotients Regression Technique (QRT) for adapting empirical models, analytically compare the QRT with existing adaptation techniques, and to also perform a path loss prediction comparison of the QRT adapted empirical models with artificial neural network (ANN) based models. The objectives of the study are as follows:
i) Determination of reliabilities of empirical models for path loss prediction across the terrains under investigation.
12
ii) Development of the QRT for empirical model adaptation.
iii) Determination the accuracy of the QRT relative to existing techniques.
iv) Determination of path loss prediction accuracy of QRT adapted empirical models relative to empirical models adapted by other techniques.
v) Creation of the ANN-based prediction models.
vi) Determination of path loss prediction accuracy of QRT adapted empirical models relative to ANN-based prediction models.
vii) Determination of the most suitable path loss prediction models for the terrains under investigation.
1.3 Statement of Problem Adequate knowledge of radio propagation characteristics across a specific terrain is an essential requirement in the planning of a wireless telecommunications network. Since path loss serves as the dominant factor for the characterization of a radio link, there is need for accurate path loss prediction so that the radio path can be optimally engineered. Path loss varies from one environment to the other according to the physical nature, dimensions and geometries of the various obstacles that perturb radio propagation. Hence, it is of high necessity to create prediction models that are not only very accurate, but also computationally efficient.
Although empirical models are quite simple to implement, they are not universally applicable due to terrain diversity across the globe, in spite of the availability of correction factors. The most popular technique for adapting empirical models has to do with introducing computed errors as correction factors into empirical model expressions. However, these correction factors in most cases only modify the constant within an expression, disregarding the
13
slope coefficient, which is a dominant factor in determining how well an empirical model fits (or is adapted to a given terrain). By implication, if the slope of the best fit curve through measured path loss points significantly differs from that of the empirical model expression, such a technique will be highly inaccurate. As a result, these techniques are limited in terms of ability to accurately adapt empirical models to terrains due to terrain diversities. Hence, it is necessary to develop a technique that does not only provide the best possible fit for the empirical model, but also ensures that the adapted empirical model is robust when tested with new data. In this study, a novel technique for adapting empirical models, termed the Quotients Regression Technique (QRT) is proposed. The QRT provides the best possible fit for an empirical model by directly fitting the empirical model onto the best fit curve for measured path loss points, thereby ensuring greater correlation with the measured data. Existing literature have revealed that computational intelligence techniques are the recent alternative approaches used to predict the path loss at a particular location in an investigated area (Ostlin, 2010). Such techniques include Artificial Neural Networks (ANNs). ANNs have the ability to handle non-linear function approximation with greater accuracy than those techniques which are based on linear regression. Hence, further in this study, QRT adapted empirical models are analytically compared for path loss prediction accuracy with ANN-based models, as well as with empirical models adapted by existing techniques. As case study, at an operating frequency of 900MHz, three Nigerian terrains are considered: the rural area between Jos and Abuja, the urban terrain (Abuja), and the semi-urban terrain (Maiduguri).
14
1.4 Methodology The methodology adopted consists of the following series of activity.
i) Received power measurement and path loss Computation.
ii) Determination of reliabilities of empirical models for path loss prediction within the terrains under investigation.
iii) Development of the Quotients Regression Technique (QRT) for empirical model adaptation.
iv) Adaptation accuracy comparison of the QRT with the Okumura GAREA (Gain due to type of environment) Technique and the RMSE (Root Mean Squared Error) Adaptation Technique (RAT). The Okumura GAREA technique involves adapting the Okumura model to a given rural terrain using a GAREA value obtained from the Okumura Curves, as correction factor. On the other hand, the RAT has to with either subtracting a computed RMSE from the model expression if path loss is overestimated, or vice versa.
v) Performance Comparison of the QRT adapted empirical models with the Okumura GAREA and RAT adapted empirical models.
vi) Creating the artificial neural network predictors. The two types of Feed-forward ANNs considered are the Multilayer Perceptron Neural Network (MLP-NN) and the Radial Basis Function Neural Network (RBF-NN).
vii) Comparative analysis of QRT adapted empirical models with ANN based predictors.
15
1.5 Significance of the Study As wireless telecommunication becomes ever increasingly popular and indispensable, it is of paramount importance to ensure quality delivery of service to subscribers within a service area. This is achievable through proper network planning and optimization. Radio propagation models are widely used in coverage determination during network planning. A model that satisfies a given terrain may not be suitable to a different terrain due to clutter differences across the globe. Hence, it is necessary to come up with efficient and effective techniques for formulating terrain – specific radio propagation models that can accurately predict path loss within a given area. This is what this study aims to achieve. 1.6 Thesis Outline This thesis consists of five chapters. The introductory part comprising of the background, aim and objectives, Statement of Problem, methodology, as well as the thesis outline are presented in chapter one. The review of fundamental concepts and review of similar works are contained in chapter two. Chapter three comprises of the materials and methods which describe the measurements procedure, path loss computation, development of the proposed Quotients Regression Technique (QRT), empirical model adaptation, neural network models creation and the comparative analysis of QRT adapted empirical models with neural network based models for path loss prediction. Results and discussions of empirical model adaptation, comparison of adaptation techniques, as well as the comparison of QRT adapted empirical with neural network based models, are presented in Chapter four. Chapter 5 contains summary, significant contributions, conclusions, recommendations for further works, as well as limitations
IF YOU CAN'T FIND YOUR TOPIC, CLICK HERE TO HIRE A WRITER»
