On the Optimization of geolocation Problems in wireless sensor Networks and connected Objects

Université de Monastir Universidad de Sevilla Doctoral School National Engineering School of Monastir University Department Electrical Departement Thesis defended by Rahma Mani Defended on November 30, 2024 In order to become Doctor from Université de Monastir and from Universidad de Sevilla Academic Field Electrical Engineering Speciality Computer Engineering On the optimization of geolocation problems in wireless sensor networks and connected objects Thesis supervised by Pr. Noureddine Liouane Pr. Antonio Rios Navarro Committee members Referees Pr. Jaime Lloret Mauri Pr. Ali Douik Examiners Pr. Lotfi Nabli Committee President Pr. Jose Luis Sevillano Ramos Supervisors Pr. Noureddine Liouane Pr. Antonio Rios Navarro Colophon Doctoral dissertation entitled “On the optimization of geolocation problems in wireless sensor networks and connected objects”, written by Rahma Mani, completed on November 18, 2024, typeset with the document preparation system L A T E X and the yathesis class dedicated to theses prepared in France. Université de Monastir Universidad de Sevilla Thesis defended by Rahma Mani Defended on November 30, 2024 In order to become Doctor from Université de Monastir and from Universidad de Sevilla Academic Field Electrical Engineering Speciality Computer Engineering On the optimization of geolocation problems in wireless sensor networks and connected objects Thesis supervised by Pr. Noureddine Liouane Pr. Antonio Rios Navarro Committee members Referees Pr. Jaime Lloret Mauri Pr. Ali Douik Examiners Pr. Lotfi Nabli Committee President Pr. Jose Luis Sevillano Ramos Supervisors Pr. Noureddine Liouane Pr. Antonio Rios Navarro Doctoral School National Engineering School of Monastir University Department Electrical Departement Université de Monastir Universidad de Sevilla École doctorale National Engineering School of Monastir Unité de recherche Electrical Departement Thèse présentée par Rahma Mani Soutenue le 30 novembre 2024 En vue de l’obtention du grade de docteur de l’Université de Monastir et de l’Universidad de Sevilla Discipline Génie Électrique Spécialité ingénierie informatique Sur l’optimisation des problèmes de la géolocalisation dans les réseaux de capteurs sans fil et les objets connectés Thèse dirigée par Pr. Noureddine Liouane Pr. Antonio Rios Navarro Composition du jury Rapporteurs Pr. Jaime Lloret Mauri Pr. Ali Douik Examinateurs Pr. Lotfi Nabli président du jury Pr. Jose Luis Sevillano Ramos Directeurs de thèse Pr. Noureddine Liouane Pr. Antonio Rios Navarro ACKNOWLEDGMENTS This thesis was carried out under a cotutelle agreement between the University of Seville, Spain, and the National Engineering School of Monastir (ENIM), University of Monastir, Tunisia. This research work is carried out within the computer architecture and technology Department of the Higher Technical School of Computer Engineering (ETSII) in collaboration with the Laboratory for Research in Automation, Signal and Image Processing "LARATSI" located at the National School of Engineers of Monastir (ENIM). I would like to express my sincere gratitude to my thesis directors Professeur NOUREDDINE LIOUANE and JOSE ANTONIO RIOS NAVARRO, for their advice, their scientific supervision and their exceptional availability. Their advice and patience were important elements which characterized the fruitful cooperation between the University of Seville and the University of Monastir during these years of thesis. I offer my sincere thanks to the members of the jury for the honor they have conferred on me by taking the time to read and evaluate the work contained in this thesis. Finally, my warmest thanks go to my family and my loved ones for their love and support throughout these years of thesis. v CONTENTS Acknowledgments v Contents vi List of Tables ix List of Figures x Abstract xii 1 General Introduction 1 1.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 1.2 Problem formulation and Motivations . . . . . . . . . . . . . . . 2 1.3 Objectives and scope of the report . . . . . . . . . . . . . . . . . . 2 1.4 Thesis structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 1.5 Publications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4 2 Context and state of the art: Wireless Sensor Networks and Localization 5 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5 2.2 Wireless sensor networks architecture . . . . . . . . . . . . . . . 6 2.3 Wireless sensor node anatomy . . . . . . . . . . . . . . . . . . . . 8 2.3.1 Detection unit . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3.2 Processing unit . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.3.3 Communication unit . . . . . . . . . . . . . . . . . . . . . . 9 2.3.4 Unit of energy . . . . . . . . . . . . . . . . . . . . . . . . . 9 2.4 Wireless sensor network applications . . . . . . . . . . . . . . . . 9 2.4.1 Military applications . . . . . . . . . . . . . . . . . . . . . . 10 2.4.2 Medical applications . . . . . . . . . . . . . . . . . . . . . . 10 2.4.3 Agricultural applications . . . . . . . . . . . . . . . . . . . 11 2.4.4 Smart-home applications . . . . . . . . . . . . . . . . . . . 12 2.4.5 Environmental applications . . . . . . . . . . . . . . . . . . 12 2.4.6 Industry 4.0 applications . . . . . . . . . . . . . . . . . . . 12 2.5 Wireless sensor network challenges . . . . . . . . . . . . . . . . . 13 2.6 Importance of localization in WSNs . . . . . . . . . . . . . . . . . 14 2.7 Distance measurment techniques classification . . . . . . . . . . 15 2.7.1 Range-based algorithms . . . . . . . . . . . . . . . . . . . . 15 2.7.2 Range-free algorithms . . . . . . . . . . . . . . . . . . . . . 17 2.8 General localization algorithms . . . . . . . . . . . . . . . . . . . 21 2.8.1 Trilateration . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 vi Contents vii 2.8.2 Multilateration . . . . . . . . . . . . . . . . . . . . . . . . . 22 2.8.3 Triangulation . . . . . . . . . . . . . . . . . . . . . . . . . . 23 2.8.4 Centroid . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24 2.8.5 Max-Min method . . . . . . . . . . . . . . . . . . . . . . . . 26 2.8.6 Fingerprint method . . . . . . . . . . . . . . . . . . . . . . 27 2.9 Machine Learning and localisation in WSNs . . . . . . . . . . . . 27 2.10 Evaluation criteria for a localization algorithm . . . . . . . . . . . 30 2.11 The Mobile WSN and path-planning . . . . . . . . . . . . . . . . 31 2.12 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33 3 Fuzzy machine learning for localization in Range-Free WSNs 35 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35 3.2 Soft computing techniques for WSN localizations . . . . . . . . . 36 3.3 Fuzzy logic and Extreme Learning Machine concepts . . . . . . . 39 3.3.1 Fuzzy logic concept for localization in WSN . . . . . . . . 39 3.3.2 ELM concept for localization in WSN . . . . . . . . . . . . 41 3.4 Geolocalization via Lightweight Fuzzy-ELM . . . . . . . . . . . . 44 3.4.1 Fuzzy-ELM simulation and statistical results . . . . . . . . 50 3.4.2 Experimental analysis of positioning . . . . . . . . . . . . 50 3.4.3 Influence of the variation of the number of nodes . . . . . 51 3.4.4 Effect of communication range . . . . . . . . . . . . . . . . 52 3.4.5 Robustness analysis for irregular connectivity . . . . . . . . 53 3.5 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55 4 Mobile Bounding-Box for Localization in WSN 56 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 4.2 Large-scale WSNs localization algorithms . . . . . . . . . . . . . 57 4.3 The basics of Kalman-Filter . . . . . . . . . . . . . . . . . . . . . 59 4.4 The proposed localization algorithm: BBLA-KF . . . . . . . . . . 60 4.4.1 Mobility method . . . . . . . . . . . . . . . . . . . . . . . . 61 4.4.2 Mobile 3D Bounding Box algorithm 3D-MBBX . . . . . . . 62 4.4.3 Kalman filter refinement process . . . . . . . . . . . . . . . 64 4.5 BBLA-KF simulation results . . . . . . . . . . . . . . . . . . . . . 66 4.5.1 Experimental analysis of positioning . . . . . . . . . . . . . 67 4.5.2 Comparative analysis with other localization techniques . . 69 4.5.3 Influence of the variation of the number of nodes . . . . . . 70 4.5.4 Influence of anchor node positions ratio . . . . . . . . . . . 71 4.5.5 Effect of communication range . . . . . . . . . . . . . . . . 72 4.5.6 Effect of anisotropy . . . . . . . . . . . . . . . . . . . . . . 73 4.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 5 Regularized BBLA-KF localization algorithm and Hardware implementation 76 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 5.2 Hardware implementation and 3D Mobile WSN localization . . . 79 5.3 Regularized BBLA-KF algorithm for localization in WSN . . . . . 82 5.4 Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5.4.1 Regularization factors effects . . . . . . . . . . . . . . . . . 84 5.4.2 Qualitative performances . . . . . . . . . . . . . . . . . . . 87 Contents viii 5.5 FPGA implementation of the RBBLA-KF algorithm . . . . . . . . 91 5.5.1 Work methodology and tools . . . . . . . . . . . . . . . . . 92 5.5.2 FPGA implementation . . . . . . . . . . . . . . . . . . . . . 92 5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99 Conclusions and future prospects 100 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Thesis contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 Future prospects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 Conclusiones y perspectivas 102 Conclusiones . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Contribuciones a la tesis . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Perspectivas de futuro . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Bibliography 104 Glossary 114 LIST OF TABLES 2.1 Range-based versus Range-free methods. . . . . . . . . . . . . . . . 20 3.1 Exampleoffunctions .......................... 42 3.2 correspondence between HCand their positions XY ......... 46 3.3 Fuzzyrules................................ 46 3.4 ALE results for different DOI values. . . . . . . . . . . . . . . . . . 55 4.1 Localization error results of hundred trials . . . . . . . . . . . . . 69 4.2 Simulation parameters of nodes number variation . . . . . . . . . 70 4.3 Localization error CDF’s for different number of nodes . . . . . . 71 4.4 ALE results for isotropic and anisotropic 3D fields . . . . . . . . . 73 5.1 Simulation parameters . . . . . . . . . . . . . . . . . . . . . . . . . 85 5.2 αzvariation (αy= 0.5; αx=0.5)..................... 85 5.3 αyvariation (αz= 1; αx= 0.5)...................... 85 5.4 αxvariation (αz= 1; αy= 0.75)..................... 85 5.5 Statistical results for different number of nodes . . . . . . . . . . . 89 5.6 Statistical results for different values of communication range . . 91 5.7 The required resources for implementing our proposed method on the Zynq board for floating point format . . . . . . . . . . . . . . . 96 5.8 The required resources for implementing our proposed method on the Zynq board for fixed point values . . . . . . . . . . . . . . . . . 97 5.9 Performances for fixed-point and floating-point FPGA implementations.................................... 97 5.10 System Generator results versus MATLAB results . . . . . . . . . . 98 ix 1.4. Thesis structure 3 applications. We have proposed Fuzzy-ELM algorithm which is given to improve localization accuracy in WSNs in the Range-Free case. The Extreme Learning Machine (ELM) and the fuzzy-logic idea are two artificial intelligence techniques that are utilized in the WSN localization process. Also, we have proposed a localization algorithm, Regularized Bounding-Box within the Kalman Filter algorithm (RBBLA-KF), adapted to networks with multi-hop communication and high deployment density based on a recursive aspect, with the aim of efficiently estimating the positions of nodes in their geographical area. The algorithm is an improvement of the original algorithm (BBLA-KF), utilizing a combination of a Kalman filter and bounding box localization algorithm. Moreover, The localization algorithm was implemented on the FPGA board “Zynq MiniZed 7007s evaluation board” and was evaluated by comparing the performances achieved through two different implementations: Hardware and Software implementation. 1.4 Thesis structure The manuscript is organized into four chapters as follows: The first chapter presents the state of the art, offering a bibliographical review, basic elements, and general concepts of wireless sensor networks, focusing on their applications in different fields especially the Internet of Things (IoT) for military purposes and in Industry 4.0. It also discusses the various methods for solving geolocation problems. Range-based and range-free distance measurment techniques, as well as various algorithmic approaches for localization, are presented in detail. The second chapter focuses on the application of artificial neural networks within static wireless sensor networks. It then presents neural network architectures and learning algorithms, as well as their variants, and discusses the Extreme Learning Machine (ELM) method. The operation of the ELM is detailed, as are the corresponding learning algorithms. We detail our contribution, which consists of a Fuzzy-ELM algorithm for WSNs, notably our Fuzzy-ELM method composed of fuzzy logic and ELM. The simulation results, discussed at the end of the chapter, demonstrate the capability of our algorithm compared to conventional localization heuristics in terms of localization accuracy for large scale static WSNs. In the third chapter, we present our second contribution to this thesis, which consists of an optimized localization algorithm in a mobile wireless sensor network. A detailed description of the localization approach using the mobile bounding box and Kalman filter for geolocation in mobile wireless sensor networks is presented. This chapter also includes an analysis of simulation results for isotropic and anisotropic networks. We prove that our proposed localization algorithm (BBLAKF) has a high accuracy in estimating the positions of sensor nodes in the network 1.5. Publications 4 compared to other localization algorithms in the literature. In the fourth chapter, a prototype implementation structure for our RBBLA-KF algorithm on FPGA hardware architecture is presented, using the FPGA simulation and implementation tools “Matlab-XSG” under the Simulink environment. Synthesis and performance evaluation are carried out using High-Level synthesis tools (Vivado-HLS). Finally, this manuscript closes with conclusions and perspectives presented in chapter 5. 1.5 Publications This thesis has given rise to two international scientific paper publications and two international conferences publications listed below: • Improved 3D localization algorithm for large scale wireless sensor networks.[1] • Localizing unknown nodes with an FPGA-enhanced edge computing UAV in wireless sensor networks: Implementation and evaluation. [2] • Improved Least-Square DV-Hop Algorithm for Localization in large scale wireless sensor network [3] • Improved Distance vector based Kalman Filter localization algorithm for wireless sensor network [4] CHAPTER TWO CONTEXT AND STATE OF THE ART: WIRELESS SENSOR NETWORKS AND LOCALIZATION 2.1 Introduction Wireless sensor networks (WSNs) are becoming increasingly relevant, largely due to recent advances in hardware and communication. In particular, the size, computing performance, and energy efficiency of the sensing devices are improving at a high rate while the price of these devices is constantly decreasing. Adding to this, services built on WSN are available to a wide range of users, thus the need for efficient management and smoothing of the data gathered by these nodes increases. In this respect, accurate localization is crucial for providing services or applications that rely on efficient data gathering or processing. Therefore, it is important to improve the positioning accuracy of sensor nodes as much as possible by developing particularly robust localization algorithms. On the one hand, the objective of this chapter is to briefly present the general concepts of wireless sensor networks. The various WSN constraints are also discussed. On the other hand, an in-depth study of the geolocation problem in wireless sensor networks is presented. In addition, we will provide a classification of the different localization approaches that exploit "range-free" and "range-based" localization algorithms. We will also present some examples of these algorithms and show their advantages and disadvantages. 5 2.2. Wireless sensor networks architecture 6 2.2 Wireless sensor networks architecture Recent advances in IOT and Industry 4.0 (IIOT) applications have led to the emergence of a new trend in network structure, known as Wirless Sensors Network (WSN). This new technology has changed the behaviour and operation of many current embedded system applications, thanks to its low cost and wireless communication support. The WSN is becoming a very popular technology that is attracting a great deal of research interest. Figure 2.1 shows an example of the operation of a wireless sensor network architecture. Figure 2.1: Example of the architecture of a wireless sensor network (WSN) Figure 2.2: WSN structure WSNs consist of a large number of small devices, called sensor nodes (Figure 2.2), deployed over a wide area of interest, which capture and communicate information from its sensing environment, such as temperature, humidity, sound, 2.2. Wireless sensor networks architecture 7 Figure 2.3: WSN categories light, and motion, and transmit it to a base station, called a Sink, through radio frequency signals via transmitters/receivers which in turn retransmits the collected data to the end user. The terms "anchor nodes" and "unknown nodes" refer to nodes with known and unknown locations, respectively. The sensor nodes can be fixed or mobile within the network (Figure 2.3). Mobility for sensor networks is a new challenge where each node can mobilize within the network, alone or with a group of nodes. In such applications, sensors are embedded in mobile devices such as mobile robots, helicopters, vehicles, or animals. Deployment consists of setting up the network nodes by human intervention. This deployment can be either deterministic, assigning well-defined positions to the sensors, or random. In the case of random deployment, non-deterministic deployment, the sensors auto-organize to form the network, so their positions are not known in advance. The latest developments in hardware design have made it possible to develop fully functional sensor network devices that are small in size, with limited battery life and computing capacity, but which integrate an interesting variety of sensors and communication technologies. These low-cost and low-power devices are capable of collecting, processing, and communicating data using low-bandwidth radio links. WSNs are capable of detecting a wide range of parameters with high accuracy and low energy consumption. In fact, many sensor nodes are widely used in geographical areas, some of which have to work in harsh environments for long periods of time. Energy conservation is analyzed as the most critical issue in the design and operation of WSNs, due to the impact on overall network lifetime and efficiency. The different ways of achieving energy savings can be linked to the optimisation of the following parameters: • a: The number of transmissions required can be minimised. • b: Optimised routing paths can be used in networks. 2.3. Wireless sensor node anatomy 8 • d: Processing algorithms with low algorithmic complexity. • c : Microcontrollers or FPGAs with low energy costs. • e: Energy recovery systems. WSNs are capable of integrating into the Internet of Things environment via router or switcher gateways. Currently, WSN implementation involves two basic uses of nodes, which are single sensor nodes and multiple sensor nodes. A single sensor node can be designed using an integrated sensor unit, signal conditioner and network application processor. Multi-sensor nodes are used to acquire a variety of physical quantities with a single wireless sensor node. 2.3 Wireless sensor node anatomy A wireless sensor node is a microsystem responsible for monitoring a physical event, collecting data related to the monitored event and sending it to the base station via radio communication links. A sensor node is usually equipped with a radio transceiver, a microcontroller, a small amount of memory, sensors, and a power supply. A wireless sensor node is often made up of four basic units: the sensor unit, the processing unit, the transmission unit and the energy unit. Depending on its field of application, it may also include other components, such as a GPS localization system or another localization system capable of determining the position of a node and a mobilisation system capable of moving the sensor node to different positions if required. Figure 2.4 shows the architecture of a sensor node. The various basic units are described in the sections below. Figure 2.4: WSN sensor node architecture 2.4. Wireless sensor network applications 9 2.3.1 Detection unit The detection unit, generally consisting of the sensor itself and an analogue- to-digital converter (ADC). The sensor transforms the physical quantity into an electrical quantity that is an image of the parameter to be processed. The ADC converts the analogue value into a digital value that can be used by the node’s processing unit. 2.3.2 Processing unit The Processing unit is made up of two modules: a microcontroller and/or a low-power FPGA and a second module consisting of a memory area. The role of the processing unit is to execute procedures enabling the sensor nodes to work together or to store the data collected in a RAM and to perform acquisition and pre-processing tasks. 2.3.3 Communication unit The communication unit representing the radio frequency transmitter / receiver that ensures wireless communication between the various wireless sensor nodes. 2.3.4 Unit of energy The energy source that supplies power to the various units mentioned to ensure that the whole wireless sensor functions correctly. In most applications, this battery is neither rechargeable nor replaceable. It is therefore the main constraint when designing RF modules and wireless sensor network communication protocols. In fact, the battery has an energy capacity that is limited in time and is generally non-rechargeable. 2.4 Wireless sensor network applications Wireless sensor networks have attracted global research interest in the scientific community and industrial applications, due to their broad applications such as smart homes, and environmental monitoring, healthcare (Figure 2.5), military surveillance, etc. Many of these applications are tailored to specific regions. For example, given that the forest is considered one of the most important and vital natural resources and also plays an important role in the ecological balance of the planet, then a WSN can be used to monitor forest resources in order to 2.4. Wireless sensor network applications 10 avoid and prevent threats. In fact, early detection of a threat in a forest could significantly reduce the damaged area and preserve wildlife. WSNs can also be used to monitor agricultural fields. In fact, it can detect nutrient deficiencies, disease, lack or excess of water, the presence of weed infestations, insect damage, wind damage and so on. Consequently, by using a WSN, the farmer can monitor air and soil temperature, humidity, CO2 concentration, and receive warnings about potential threats to crop fields. Reference [5] presents dedicated WSN applications for weather monitoring and disaster alarm systems. Reference [6] also presents applications in air pollution monitoring systems. References [5][2] provide a comprehensive review of the literature on WSN applications, including dedicated military and surveillance uses. Several projects have been developed for military surveillance in battlefields and national protection against violence and attacks. Notably, the DSN (Distributed Sensor Network) project, conducted by DARPA (Defense Advanced Research Projects Agency) in the 1980s, is a prime example [7]. WSNs can also be applied in the automotive and robotics fields. The development of wireless communication systems, the miniaturised size and low cost of sensors, combined with their ability to collect different kinds of information (temperature, pressure, humidity, vibration, etc.) have increased the range of applications for WSNs. In the sections below, we describe the best-known application areas for WSNs. 2.4.1 Military applications The military sector was one of the first applications that led to the use of WSNs. These networks offer a number of advantages, including rapid deployment, fault tolerance, low cost and the ability to self-organise without a fixed infrastructure. All these advantages have made WSN a desirable tool in military applications. A sensor network placed in a strategic or inaccessible region is responsible for detecting all enemy movements or analysing the territory (detection of nuclear bombs, chemical or biological agents, radiation, etc) before deploying forces. In this way, sensor networks can be used to guide military units through a battle zone or to prevent intrusions and terrorist attacks in cities. Several projects dedicated to WSNs have been developed in this context. One example is the DSN (Distributed Sensor Network) [7] SensIT (Sensor Information Technology) [8] project, which was led by the US-DARPA. 2.4.2 Medical applications WSNs are used in the medical field to observe the physiological state of a vital organism in real time. Capsules in the form of micro-sensors or micro-cameras can 2.4. Wireless sensor network applications 11 be implanted under the skin to deliver images inside the human body without the need for surgery. Other biomedical applications involving the use of WSNs today include blood sugar control, organ monitoring and early cancer detection. This wireless technology also enables doctors to diagnose patients remotely (see figure 2.5). WSNs are used in the medical field to observe the physiological state of a vital organism in real time. Capsules in the form of micro-sensors or micro-cameras can be implanted under the skin to deliver images inside the human body without the need for surgery. Other biomedical applications involving the use of WSNs today include blood sugar control, organ monitoring and early cancer detection. This wireless technology also enables doctors to diagnose patients remotely (see figure 2.5). WSNs are used in the medical field to observe the physiological state of a vital organism in real time. Capsules in the form of micro-sensors or micro-cameras can be implanted under the skin to deliver images inside the human body without the need for surgery. Other biomedical applications involving the use of WSNs today include blood sugar control, organ monitoring and early cancer detection. This wireless technology also enables doctors to diagnose patients remotely (see figure 2.5). In addition, these networks can detect abnormal behaviour (falling Figure 2.5: Example of healthcare application on WSN out of bed, shock, screaming, etc.) in disabled or elderly people [9]. 2.4.3 Agricultural applications The integration of WSNs in agricultural applications would make it possible to assess well-defined parameters such as moisture content, temperature and so on. The parameters captured are aggregated and generally sent to a processing centre to help farmers optimise water or nutrient irrigation. WSNs can also be deployed to detect dry areas in a crop field, hence the term ‘intelligent agriculture’ or ‘IOT agriculture’. The use of precision technology gives farmers greater con- 2.4. Wireless sensor network applications 12 trol over their processes. It gives them real-time information on measurements that are essential for optimised management of growing areas. WSNs offer the possibility of real-time monitoring of large geographical areas that are generally free of infrastructure. The data captured by these networks can be processed and transmitted to an end user for further use. 2.4.4 Smart-home applications Wireless technology allows home appliances to be controlled locally via a wireless sensor network or remotely via IOT concepts. Sensor nodes can be integrated into ovens, microwaves, fridges and video recorders, simplifying the task of controlling these devices for users. Smart homes constitute an autonomous environment, capable of analysing situations through the behaviour of people, in order to make effective decisions in real time. Also, the wireless sensors built into homes can be used to ensure the occupant’s safety. 2.4.5 Environmental applications The implementation of WSNs in environmental applications enables environmental parameters to be monitored in real time [10]. We can find sensor nodes in forests to warn of possible fires and then plan effective fire-fighting solutions in real time. WSNs can also be integrated into urban environments to measure environmental parameters in collection zones, and detect potential environmental threats such as pollution, meteorological hazards, etc. 2.4.6 Industry 4.0 applications Industry 4.0, also known as connected industry through the Internet of Things, was invented in 2011 as an initiative of the German government. It refers to the fourth generation of industry, that of today. Industry 4.0 focusses primarily on the exploitation of information and telecommunication technologies in the manufacturing industry at the production stage. In other words, Industry 4.0 is the connected informatization of manufacturing (Figure 2.6). Wireless sensors and wireless sensor networks offer considerable advantages in this context as they are flexible and easy to deploy. They reduce installation and maintenance costs, and help reduce the need for pre-installed electrical cables and infrastructure, thus improving the overall productivity of the production system. WSNs can be installed on machines to provide direct information from inside the process. In addition, during the data analysis phase, it is preferable to know the location 2.7. Distance measurment techniques classification 19 Figure 2.8: The entire flowchart of the traditional DV-Hop algorithm mum number of hop-count between node k and anchor i . Step 3, Once the average distance of hops is known, the determination of the unknown node position is launched using the multilateration method. The general entire flowchart of the traditional DV-Hop algorithm is described in Figure 2.8. Error Analysis of the Original DV-Hop Localization Algorithm In this step, the error of localization is caused owing to the calculation of the average distance per hop, hop-size, by each anchor. Citing an example illustrated in Figure 2.9 of WSN network topology containing three anchor nodes A1 , A2 , A3 , and unknown nodes Ui where i from 1 to N . After the first step, each anchor node is aware of its location and its minimum hops from each unknown node. The blue line connecting two nodes notify the direct communication transfers between them. Therefore, the communication range of each sensor is 10 m and the distances between anchors are 20, 30, 40 m from A3 to A1 , A3 to A2 , and A1 to A2 respectively. Also, 3, 5, 4 are the number of hops between A1 : A3 , A1 : A2 , and A3 : A2 respectively. The average hop-size of the three anchors in this network is calculated as follows using the original DV-Hop algorithm, Equation(2.2). 2.7. Distance measurment techniques classification 20 HA1=40+20 3+5 = 7.5 (2.4) HA2=40+30 5+4 = 7.777 (2.5) HA3=20+30 3+4 = 7.142 (2.6) As well, the number of hops from UN to A1 , A2 , A3 is 2, 3, 1 respectively. Then, the distance estimated by the Dv-Hop algorithm between the unknown node UN and the anchor A3 is 1 × 7.14=7.14m using Equation (2.3) . However, in the outline, we can notice that the real distance between A3 and the unknown node is 10 which proves the error made by the method of calculating the average of hop sizes by anchors and to demonstrate how far the results are. Figure 2.9: Node distrubition model Compared with Range-based algorithms, Range-free algorithms do not require additional materials for localisation. Furthermore, Range-free algorithms are cost-effective and energy-conserving solutions for WSNs. So, in this work, we are interested in Range-free localisation algorithms. Table 2.1 shows a brief comparison between Range-based and Range-free methods. Range-based Range-free High precision Less precision Additional equipment required No need for extra equipment Sensitivity to environmental noise More robustness Table 2.1: Range-based versus Range-free methods. 2.8. General localization algorithms 21 2.8 General localization algorithms Existing localization algorithms can be presented in a variety of ways: distributed algorithms, centralized algorithms, iterative algorithms, mobility-assisted approaches, etc. Each category corresponds to an appropriate localization technique. The researchers propose a varied number of localization algorithms depending on the computational capacity of each sensor nodes. 2.8.1 Trilateration The trilateration method is a geometric method based on the distances separating the unknown node to be located from the various anchors, as well as the geometric coordinates of these anchors, which must not be aligned. Given three anchors A1 , A2 and A3 , with coordinates ( x1 ; y1 ) , ( x2 ; y2 ) and ( x3 ; y3 ) respectively, U the unknown node with coordinates ( x ; y ) to be estimated. After execution of the distance measurement protocol, a triplet ( d1 ; d2 ; d3 ) has been deduced. The distance data will later be broadcast in the network, allowing the unknown node U to know its position. The relationship between anchors A1 , A2 and U is given by the following equations: (x−x1)2+(y−y1)2=d2 1(2.7) (x−x2)2+(y−y2)2=d2 2(2.8) The desired position is therefore at the intersection of circles C1 ( A 1; d 1) and C2 ( A 2; d 2) gives the desired position. Generally, circles C1 and C2 have two distinct points of intersection P and P′ . Using the data from anchor A 3, the node’s position will be chosen between these two points of intersection, as shown in figure 2.10. Figure 2.10: Example of trilateration technique 2.8. General localization algorithms 22 2.8.2 Multilateration The multilateration method presents the case where the number of anchors is greater than three, resulting in an over-dimensioned system, in other words, where the number of equations is greater than the number of unknown variables. The set of the equations presented by the following system formulates the multilateration method:                        (x1−xk)2+(y1−yk)2=d2 1k (x2−xk)2+(y2−yk)2=d2 2k ... (xi−xk)2+(yi−yk)2=d2 ik (xN−xk)2+(yN−yk)2=d2 Nk (2.9) where ( xi,yi ) ( i = 1,2,... N ) is the location of anchor node i , dik is the valued distance segregating an anchor i from the unknown node k . The expansion of the Equation (2.9) done by the subtraction of the ( N− 1) equations from the last equation of the system conducts for a system containing N− 1 equations. Then, converting this system of equations into a matrix form defined as follows AX =B(2.10) We designate the X=(xk,yk), And A= 2                 (x1−xN) (y1−yN) (x2−xN) (y2−yN) .. .. (xN−1−xN) (yN−1−yN)                  (2.11) B=                 x2 1−x2 N+y2 1−y2 N+d2 Nk −d2 1k x2 2−x2 N+y2 2−y2 N+d2 Nk −d2 2k .. x2 N−1−x2 N+y2 N−1−y2 N+d2 Nk −d2 (N−1)k)                  (2.12) The matrix A encodes geographic information about the deployment of anchor nodes, the vector B gives information about the measured distances between sensor nodes, and X presents the unknown positions of the sensor nodes to be estimated. Finding the location of the unknown node k could be calculated by solving the equation (2.10) using the least square method. The following expression is presented the solution X= (ATA)−1ATB(2.13) 2.8. General localization algorithms 23 Figure 2.11: Example of Multilateration technique 2.8.3 Triangulation Triangulation is a trigonometric method for determining the position of a point by forming a triangle from known reference points and measuring the angles between this point and these reference points. The unknown node can estimate its position from the angle of incidence of its transmitted signal. Let’s consider two anchors A1 ( x1 ; y1 ) and A2 ( x2 ; y2 ) communicating with an unknown node U ( x ; y ). Anchors A1 and A2 have equipment for measuring the angles at which they receive the signal emitted by node U . Figure 2.12 and equations (2.14, 2.15) show how to estimate the position of the uknown node U. tan(α1) = y1−y x1−x(2.14) tan(α2) = y2−y x2−x(2.15) By rearranging the equations 2.14 and 2.15, we can get the solution in the form of the following matrix:       tan(α1)−1 tan(α2)−1           x y     =      x1tan(α1)−y1 x2tan(α2)−y2     (2.16) Afterward, converting the former matrix into an equation is defined as follows: ΘX=Ω(2.17) The equation 2.17 leads us to deduce the position of any unknown node U 2.8. General localization algorithms 24 Figure 2.12: Example of triangulation technique based on the positions of anchor nodes using the following equation: X= (ΘTΘ)−1ΘTΩ(2.18) 2.8.4 Centroid The Centroid algorithm was proposed by Bulusu [28]. The fundamental principle of this technique is to consider the centroid point of neighbouring anchors as the estimated position of the unknown node. Bulusu uses the simple radio model, which is suitable for outdoor applications. Figure 2.13 shows the scenario for this localization method. This algorithm does not need to measure the distance between anchor nodes, but is based simply on the connectivity of the network. The intersection of the communication zones of each anchor forms a transmission zone. The basic operating principle is as follows: Anchor nodes periodically broadcast a set of messages in their proximity, containing their ID and location information. Unknown nodes in the network record the number of packets received. When the number of packets received exceeds a certain threshold, it can be registered as a neighboring anchor node. Finally, the unknown node adopts as its estimated coordinates the geometric centroid formed by all connected neighboring anchor nodes 2.8. General localization algorithms 25 Figure 2.13: An example of localization using the Centroid technique Each anchor periodically (period= T ) transmits a frame containing its position. It is assumed that the anchors are synchronised and there are no collisions during frame transmission. For a better understanding of the algorithm, the author uses the following terms: •R: Range of communication between nodes. •T: Time interval between two signals transmitted by an anchor. •t : Duration of time during which the unknown node collects signals from anchors, t > T . •Nsent(i;t): Number of frames sent by anchor Aiat time t. •Nrecv(i;t): Number of frames received by the unknown node at time t. •CAi: Connectivity metric of anchor Ai. •CAthresh: CA connectivity threshold. • ( xcen ; ycen ): Estimated position of the unknown node by the Centroid algorithm. • (xa;ya): Actual position of the unknown node. During the fixed time period t , the unknown node N listens to the transmission channel and collects all the signals from the anchors. Although each anchor has sent Nsent(i;t) signals, due to interference during radio propagation, the unknown node can currently receive Nrecv(i;t) signals from the anchors Ai , with Nrecv(i;t)< Nsent(i;t) . In order to determine whether an anchor actually belongs to 2.8. General localization algorithms 26 the communication range of the unknown node, the author defines the anchor connectivity metric, denoted by CAi, as follows: CAi=Nrecv(i;t) Nsent(i;t)(2.19) The value of CAi is limited by a threshold called CAthresh . There are two possible scenarios: • If CAi > CAthresh ; then the unknown node will consider the corresponding anchor Ai to be its neighbour. This node will then take the Ai anchor into account when calculating its position. • If CAi < CAthresh ; then the unknown node will understand that the anchor is not in its proximity, hence this node will not consider the anchor Ai in its position estimation. Assuming that the unknown node has k anchors A1 ; A2 ... Ak whose connectivity is greater than the CAthresh threshold, it is located in the centroid of these anchors:        xcen = (x1+x2+... +xk)/k ycen = (y1+y2+... +yk)/k (2.20) 2.8.5 Max-Min method To estimate the possible geographical positions of an unknown node, the "Max- Min" method exploits the process of intersection of squares instead of intersection of circles (Figure 2.14) . Consequently, each sensor node is assumed to be located in a zone of intersection of the squares centered by the known anchor positions. This zone of intersection is delimited by the maximum of the coordinates of the lower sides and the minimum of the coordinates of the upper sides of the resulting set of squares. The zone-limiting squares for reference nodes i , anchor Ai , is centered at the position ( xi , yi ), with 2 d side length, d being the distance taken between the unknown node and anchor Ai . Formula (2.21) can be used to find this intersection rectangle. The geographical position of the unknown node with coordinates ( xu,yu ) is estimated through the center of gravity of the intersection rectangle [29].            xu=max(xi−di)+min(xi+di) 2 yu=max(yi−di)+min(yi+di) 2 (2.21) 2.9. Machine Learning and localisation in WSNs 27 Figure 2.14: Node distrubition model 2.8.6 Fingerprint method Fingerprinting usually requires an environmental scan to obtain fingerprints or characteristics of the environment where the localization system will be used [30]. Fingerprints are generally collected in the form of RSSI or CSI measurements. Initially, various RSSI measurements are collected during an offline phase. Once the system is deployed, the online measurements (obtained in real time) are compared with the offline measurements to estimate the user’s position. Several algorithms are available to match offline measurements with online measurements, some of which are discussed below. Figure 2.15: Principle of fingerprint-based localization in WSNs 2.9 Machine Learning and localisation in WSNs Recently, thanks to machine learning methods and especially neural networks, the design and optimization of the localization system in sensor networks, as 2.9. Machine Learning and localisation in WSNs 28 well as the improvement of accuracy, have become a very active topic in recent years. Indeed, based on the correspondence between the WSN’s geographical distribution and information measured on known nodes, a cartographic geolocation model can be learned. This model is then used to estimate the position of unknown nodes. ANNs, SVMs, ELMs and Deep Learning have been applied to solve localisation problems in WSNs (Tran 2008, Zheng 2012, Cottone 2016, Javadi 2019). For example, in [31] the authors proposed a Range-Free localization scheme using an improved Support Vector Machine (SVM) method called Location SVM (LSVM). In this method, the localisation process is considered as a classification problem where the unknown nodes are defined as a set of localisation regions in 2-D space to solve the localisation problems. The classification process is built on the basis of class sets (class-X and class-Y), and each located node is assigned to class-X and class-Y according to the location parameters. In addition, the authors presented a modified version of the mass-spring optimisation algorithm (MSOA) to improve localisation accuracy with LSVM. In Javadi et al [32], the authors use the SVM machine and its Twin-SVM variant to locate sensor nodes in WSNs. Javadi considers that the Twin-SVM locates the area of the node’s expected position and uses the distributed learning algorithm for this localisation process. The average of the positions of the nodes in the detection zone can be taken as the position of the sensor node to be located. Zhang et al [12]proposed a localization method exploiting the connectivity of WSNs in the anisotropic case. The proposed technique consists of estimating the inter-node distance and selecting suitable anchors for the multilateration localisation process. This technique takes into account the size of the local jump, called the Per-Hop-Length for the anisotropic case. The authors have proposed a new distance estimation model that improves the accuracy of distance estimation in WSNs. The anchor selection procedure is based on the accuracy of the distance estimate and the geographical distribution of the anchors in order to optimise location accuracy and eliminate the bad effects resulting from inappropriately distributed anchors. In [33], the authors exploited the regularised ELM-WSN to deal with the multi-hop localisation problem. The proposed algorithm uses three steps: the detection of the training data via the correspondence between the number of hops and the physical distances separating the known nodes and the unknown nodes, and the Trilateration algorithm for the localisation process. The authors in [34] presented hybrid localisation models based on fuzzy logic and the ELM model. To optimise localisation accuracy, the PSO minimises the effects of irregular deployments, while [35] presents a fast SVM learning machine for a large-scale WSN localisation algorithm. The proposed localization algorithm transforms the WSNs location estimation position into a multi-class problem, the SVM machine for localization used is employed to solve CHAPTER THREE FUZZY MACHINE LEARNING FOR LOCALIZATION IN RANGE-FREE WSNS 3.1 Introduction Wireless sensor network applications represent one of the most recent advances in wireless communication and application areas such as IoT, Industry 4.0, e-health, e-surveillance, e-logistics and others. Wireless sensor networks are made up of a large number of low-cost, low-power radio-frequency sensors that are highly interconnected. These intelligent sensors are capable of communicating via multi-hop transmission to collect physical data and phenomena in their environment. Each device collects useful information and communicates data via WSN infrastructure. The data collected requires knowledge of the geographical location of the sensor nodes to know where the event took place and make the data collected reliable and useful. In the literature, a great deal of research has focused on the problem of geographical localization in WSNs. Indeed, particular attention has been devoted to the exploitation of machine learning techniques in the field of WSNs. Specifically, the Extreme Learning Machine (ELM) and the Fuzzy-logic techniques have been used in many WSN configurations, such as Range-Based, Range-Free, isotropic, and anisotropic environments. Fuzzy logic is a method for preparing uncertain input data by transforming it into fuzzy sets and creating rules that describe the relationships between different variables in a way that is easier for humans to understand. Combining fuzzy logic and ELM can be advantageous for dealing with complex and imprecise data, while also benefiting from the fast processing capability of ELM. The extreme learning machine concepts, as a neural network method, can be exploited as a learning 35 3.2. Soft computing techniques for WSN localizations 36 machine that adapts efficiently to the treatment of fuzzy data, making it suitable for dealing with complex, non-linear connections between uncertain variables. This combination can improve the interpretability and robustness of the model, as fuzzy logic permits vulnerabilities to be managed, whereas ELM offers effective learning and generalization capabilities. This combination finds application in a variety of fields, including intelligent control systems, pattern recognition, prediction tasks and decision support systems, where the management of uncertainty and complexity is crucial. In this chapter, we have proposed an algorithm based on the combination of the Fuzzy logic and Extreme Learning Machine concepts that we have named Fuzzy-ELM for the Range-Free case for WSN localization. The proposed Fuzzy-ELM algorithm represents a new way of dealing with the problem of machine learning in an uncertain environment. Indeed, the learning phase exploits a set of dedicated fuzzy rules to handle the localization problem and[ the least squares method in the machine learning phase]. The proposed algorithm will be tested for different scenarios to demonstrate the effectiveness of the proposed method. In the following, the machine learning techniques used for our hybrid technique Fuzzy-ELM for the localization task in WSNs will be introduced. Also, Simulation results will be presented and analyzed to compare the performance of the proposed Fuzzy-ELM with different algorithms. 3.2 Soft computing techniques for WSN localizations Localization is a key challenge for Wireless sensor networks that has a significant impact on overall network efficiency and performance. Node localization in WSNs is a critical aspect in many applications. WSN applications represent one of the most recent advances in wireless communication, industrial technologies 4.0 and IoT. Over the years, researchers have explored various techniques to address this challenge. Particularly they have focused on Dv-hop based methods and have incorporated soft computing techniques to increase localization accuracy. The Dv-hop algorithms exploit hop distances between nodes to estimate sensor node positions in a decentralized way, offering a scalable solution for large-scale WSNs. The paper [44] discusses various effective localization methods for wireless sensor networks. The methods include hop-count-based expectation of distance algorithm, Voronoi-based Monte Carlo localization algorithm, Heuristic Multidimensional Scaling, Extended Kalman Filter, Multidimensional Scaling, and Delaunay triangulation based localization scheme. Various localization methods are discussed for static and mobile sensor networks for high-accuracy positional information. The DV-Hop algorithm is a commonly used technique for localization in wireless sensor networks. Several improvements have been proposed to enhance 3.2. Soft computing techniques for WSN localizations 37 its accuracy[45][4][3]. One approach is the CMWN-DV-Hop algorithm, which introduces segmentation and weighting factors to achieve global optimization [46]. Another improvement is the OANS DV-Hop algorithm, which uses an optimum anchor nodes subset and binary particle swarm optimization to increase positioning accuracy[47]. The CADV-Hop algorithm categorizes hop counts to calculate the typical hop distance, resulting in improved localization accuracy[48]. Additionally, an improved DV-Hop algorithm modifies the shortest path table and uses hop count distribution to reduce hop distance errors, leading to higher accuracy compared to the classical DV-Hop algorithm[49]. These improvements address the shortcomings of the traditional DV-Hop algorithm and significantly reduce positioning errors[50]. Paper [51] shows The optimal CH selection for finding the shortest path among the CHs is improved by developing the hybrid K-means with Particle Swarm Optimization (PSO) based hybrid Ad-hoc On-demand Distance Vector (AODV) channeling algorithms. Alongside metaheuristic approaches[52], recent advances in soft computing have attracted attention for their effectiveness in localization tasks. Particular attention has been paid to machine learning techniques (MLT) applied to WSNs. Artificial neural network (ANN), support vector machine (SVM) and deep learning (DL) models are applied in various fields of classification, density estimation and process identification also have been applied to solve localization problems in WSNs[53][54][55][56][39]. Indeed, the MLT techniques have been used in many WSNs configurations such as Range-Based, Range-Free, in isotropic and anisotropic environments[57][45][58][59]. In the case of Range-Based WSNs, the neural network inputs have certain physical characteristics of the received signals, such as RSSI, ToA, TDoA and/or AoA, which give good localization performance, but they require additional hardware. In Range-Free WSNs techniques, based on sensor network connectivity and anchor positions, the location of unknown nodes is estimated without additional devices. The localization model neural network for Range-Free can be used for any type of isotropic or anisotropic WSN and gives acceptable accuracy[60]. In article[61], the authors categorize research dimensions in WSN primary and research domains secondary and discuss different soft computing-based clustering schemes from the past two decades to deal with energy conservation issues in sensor networks. One approach involves the use of neural network architectures, such as extreme learning machines (ELM), which can handle non-linear distortions in the propagation medium[62]. Paper[62]compares soft computing techniques for localization using fixed terrestrial anchor nodes (FTANs), aerial anchor nodes (ANs) and focuses on node localization in wireless sensor networks using unmanned aerial vehicles (UAVs) . Specifically, the use of neural network architectures like ELM which contributes to the localization of wireless sensor networks by providing a 3.2. Soft computing techniques for WSN localizations 38 highly non-linear classifier that can handle distortions in the propagation medium and improve accuracy. Also, they show that the ELM is even superior over its counterpart NN classifiers like multilayer perceptron (MLP) and radial basis function (RBF) due to its fast and strong learning ability. Soft computing techniques have shown promising results in terms of enhancing localization accuracy, network lifetime, energy consumption, and load distribution in wireless sensor networks. These techniques have made significant contributions to the localization of wireless sensor networks. Another approach is the use of fuzzy logic to rectify uncertainties in RSSI measurements, resulting in improved localization accuracy[51]. In paper[63], the authors proposed to use fuzzy logic to rectify the uncertainty in the value of RSSI which in turn results in more localization accuracy, and the proposed method leads to better per node localization accuracy and overall network localization accuracy at different node and anchor node densities as compared to the weighted centroid method. In paper[64] , a scenario based fuzzy LEACH protocol is proposed to extend the network lifetime with efficient energy consumption and by choosing the appropriate head node in each cluster based on the fuzzy parameters. Soft computing-based clustering schemes have also been employed to address energy conservation issues in sensor networks, with research focusing on fuzzy inference structures and fuzzy parameters for selecting appropriate head nodes in each cluster[61]. It’s worth noting that hybrid models have also received considerable attention from the research community; for example, In the paper [39] combined FL (Sugeno) and NN systems to achieve greater accuracy than either method in isolation. Paper [65] contributes to the localization of wireless sensor networks by using the Nero-Fuzzy method to estimate the coordinates of unspecified nodes based on the positions of anchor nodes and logic fuzzy rules. A novel neuro-fuzzy-based localisation system for WSN using node proximity is proposed in paper [66]. The authors used a neural network and fuzzy logic-based approach to estimate the location of sensor nodes based on RSSIs of anchor nodes. And the RSSI values of anchor nodes are categorized into some predefined regions and RSSI patterns are generated using fuzzy inference rules. The proposed model achieves an accuracy of up to 95% . The estimated average error rate for 120 nodes using this method equals to 2% of radio range, which is minimal in comparison to APS algorithm, since the range covered by sensor networks is broad, finding all anchor nodes is a time consuming task and needs a lot of memory. This work concentrated on the integration of extreme learning machine (ELM) and fuzzy logic methodologies. ELM, as a powerful machine learning paradigm, and fuzzy logic, as a means of managing uncertainty, have shown promise for improving the accuracy and robustness of localization methods within WSNs. In the following sections, we will refer to the existing 3.3. Fuzzy logic and Extreme Learning Machine concepts 39 literature to studythe Fuzzy logic and the evolution of ELM techniques, as well as discuss the integration of soft computing methods, in particular ELM and fuzzy logic, in the context of WSN localization. 3.3 Fuzzy logic and Extreme Learning Machine concepts 3.3.1 Fuzzy logic concept for localization in WSN In the realm of localization using artificial intelligence (AI) techniques, fuzzy logic has emerged as a significant advancement. Fuzzy logic is an effective tool for managing uncertainty and vagueness in real-world localization scenarios. Unlike traditional crisp logic, fuzzy logic allows for a more nuanced and humanlike representation of location data, considering varying degrees of certainty or ambiguity. The incorporation of fuzzy logic opens up new possibilities for realworld applications such as autonomous navigation, asset tracking, and geospatial analysis. Key components of fuzzy logic include: 1. The membership function: In traditional logic theory, an element either belongs to a set (membership = 1) or does not belong (membership = 0). In fuzzy logic, the fuzzy variable is a continuous value between 0 and 1, reflecting the degree of membership of a linguistic variables. The membership function gives a weighting coefficient to a variable, qualifying it as uncertain and characterizing its membership of a given set of linguistic variables. It is designated µ ( x ). The argument x refers to the uncertain variable. figure 3.1 shows the membership functions in the combinatorial logic and the fuzzy logic case. Figure 3.1: Membership function representation 2. Several forms are often used to describe the membership function, and the choice of form depends on the field of application and the acceptable 3.3. Fuzzy logic and Extreme Learning Machine concepts 40 degree of information loss. The most commonly used shapes in practice are Triangular, Trapezoidal and Bell-shape. 3. Linguistic Variables and Fuzzyfication: Linguistic variables are used to qualify a set of physical variables in the form of linguistic terms, depending on the type of variable to be encoded. For example, Fuzzy logic allows the use of linguistic variables (e.g., "temperature," "speed", “position”) and linguistic terms (e.g., "hot", "slow", ”near”, “far”) to describe imprecise or vague concepts. These linguistic terms are associated with fuzzy membership functions. In the fuzzyfication phase, it is necessary to encode Input/Output variables using linguistic terms, in order to a set of action and reaction rules. 4. Fuzzy Rules: Generally, several values of linguistic variables, appropriately defined by membership functions, are linked together by fuzzy rules developed by the expert. Therefore, to describe the process of reasoning and evaluating a fuzzy variable in order to draw a conclusion and perform an action, a set of rules is given. These rules are often expressed in the form of "if-then" statements and are based on fuzzy logic operators such as "and," "or," and "not." A rule is described in the following form: • IF Condition 1, AND Condition 2, ... AND condition m THEN Action. Also Rules could be expressed in this general form: • IF condition 1, THEN operation 1, OR • IF condition 2, THEN operation 2, OR ... • IF condition m, THEN operation m. 5. Fuzzy Inference: Fuzzy inference involves combining fuzzy rules and input data to draw fuzzy conclusions. It often uses methods such as fuzzy logic inference engines, like Mamdani’s or Tsugeno’s inference model, which use operations on fuzzy sets to state the output. 6. Defuzzification module: The fuzzy output needs to be converted into a crisp value for practical use after fuzzy inference. This process is called defuzzification and typically involves calculating a single value that best represents the fuzzy output. Figure 3.2 shows the conventional structure of such a fuzzy logic system. 3.3. Fuzzy logic and Extreme Learning Machine concepts 41 Figure 3.2: Conventional structure of a fuzzy logic system 3.3.2 ELM concept for localization in WSN Extreme Learning Machine is a very fast machine learning algorithm that has gained significant attention in the field of wireless sensor networks, particularly in the context of localization. ELM, known for its simplicity, efficiency, and rapid training capabilities, presents a convincing solution for tackling the challenging problem of sensor node localization. Its single-hidden-layer feedforward neural network structure, coupled with random weight initialization and analytical output weight computation, enables rapid convergence and accurate localization results. ELM’s ability to handle large-scale sensor networks, its minimal computational overhead, and its capacity to adapt to dynamic network conditions make it a valuable asset in the field of wireless sensor networks, promising enhanced location accuracy and reduced computational complexity. Given a training set [ Xi,Ti ], where Xi is the training data set and Ti is the output corresponding to Xi , the single-hidden-layer Feedforward neural network SLFNs with khidden neurons can be formulated as follows: k X 1 βk,mf(wn,xn,bk) = oj,j = 1...m (3.1) where f ( . ) is the activation function associated with the hidden layer, wi is the weighting of inputs xiand biis the bias. 3.3. Fuzzy logic and Extreme Learning Machine concepts 42 Function Formula Sigmoid function f(a,b,x)=1 1+exp(−(a·x+b)) Hyperbolic tangent function f(a,b,x)=1−exp(−(a·x+b)) 1+exp(−(a·x+b)) Radial basis function f(a,b,x)=exp(-b‖x-a‖) Multi-quadratic function f(a,b,x)=∥x−a∥+b21 2 Hard limit function f(a,b,x)=       1,a·x+b≤0 0,otherwise Cosine function f(a,b,x)=cos(a·x+b) Table 3.1: Example of functions The cost function minimizing the learning error is defined by : m X j=1   oj−tj  = 0 (3.2) In fact, the learning process attempts to find a linear relationship between the hidden layer Hand the desired output Taccording to the following equation: H·β=T(3.3) where His the matrix representing the ELM’s hidden layer. H=            f(w1·x1+b1)... f (w1·x1+bk) . . . ... . . . f(wn·xn+b1)... f (wn·xn+bk)             (3.4) In the ELM algorithm, Figure 3.3 is given as an example, after randomly fixing the hidden layer’s input weights, wn and bk , the hidden layer’s output matrix, H , can be determined. 3.3. Fuzzy logic and Extreme Learning Machine concepts 43 Figure 3.3: Structure of an ELM Learning a neural network with a single hidden layer can be transformed into solving a linear system of equations (3.3). The output weights β can be determined by : β=H†T(3.5) With H† is the pseudo-inverse matrix of H known as the Moore Penrose generalized inverse matrix [67]. ELM variants • Kernel ELM The original ELM algorithm is a neural network with a single hidden layer and a linear output layer. In the case of complex classification problems, non-linear regression and other pattern recognition tasks, ELM requires a large number of neurons in the hidden layer, resulting in a complex architecture and more appropriate computational resources. The basic Kernel Extreme Learning Machine (K-ELM) algorithm exploits so-called kernel methods and provides a learning framework better suited to regression, binary classification and multi-classification problems. In addition, Kernel ELM better manages ELM parameterization and kernel tuning, which can handle heterogeneous data sources[68] [69]. • ELM multiclass projection When dealing with high-dimensional learning problems, the generalization capability of the ELM decreases sharply, mainly due to the number of hidden layers, and learning performance deteriorates significantly. The Multiclass Projection Extreme Learning Machine (MPELM) variant transforms high-dimensional data into reasonable-dimensional data using the projection matrix obtained by singular 3.4. Geolocalization via Lightweight Fuzzy-ELM 44 value decomposition. The projection matrix can be used to assign input weights to obtain more stable learning and validation results, and only a small number of hidden neurons are exploited for the ELM operation phase [70]. • Genetic ELM Genetic ELM exploits metaheuristics to optimize ELM structure. Genetic algorithms are used to optimize the ELM structure in terms of the number of cache neurons and the ELM learning hyper-parameter. The basic idea behind ELM learning via genetic algorithms is to construct several ELM structures and combine them with crossover and mutation procedures to effectively improve the generalization performance of the ELM algorithm. Genetic-ELM encodes ELM parameters with chromosomes, representing a feasible solution to the problem. These encoding, crossover, mutation and selection procedures form the basis of the Genetic-ELM algorithm[71]. • Incremental ELM The number of hidden layers can be optimized incrementally. In the case of incremental ELM (I-ELM), only a small number of hidden neurons are initially defined at the start of the learning procedure. As learning progresses, the algorithm randomly adds new hidden neurons until it achieves a better reduction of the learning error and reaches a user-determined accuracy[72]. • On-Line Sequential-ELM OS-ELM represents the iterative version of the basic ELM exploiting the recursive version of the least squares algorithm minimizing the LS errors between the estimated vector and the target vector. In the learning phase, data are iteratively processed one by one or chunk by chunk. When new data appear, the ELM hyper-parameters are updated automatically[73]. 3.4 Geolocalization via Lightweight Fuzzy-ELM Lightweight supervised learning machine is a subset of artificial intelligence and machine learning that focuses on efficiency and compactness when developing supervised learning models. It is well suited to the field of embedded learning dedicated to IoT applications and in particular to low-power WSNs because of their energy resources and low latency. This technique explores the integration of fuzzy logic into a localization algorithm based on an ELM architecture, which is particularly beneficial in situations where obtaining accurate localization information is difficult due to environmental factors, obstacles, electromagnetic noise or 3.4. Geolocalization via Lightweight Fuzzy-ELM 51 Figure 3.8: Localization results of Fuzzy-ELM algorithm over 3D-WSN distribution 3.4.3 Influence of the variation of the number of nodes Different node densities in 3D fields are evaluated to assess the performance of our presented algorithm. The number of nodes increased from 200 to 800. The reference nodes ratio is 15% and the communication range is 40 m. The average localization error for different WSN densities can be seen in Figure 3.9 by examining the empirical cumulative distribution functions (ECDFs). In this figure, 90% of the unknown nodes can be pinpointed with an error less than 0.8 ×R for 200 nodes while 400, 600, 800 nodes give an ALE less than 0.4 ×R . We can conclude that if the number of nodes increases, the average localization error decreases thanks to the enhancement of WSN connectivity. 0 0.2 0.4 0.6 0.8 1 1.2 Average Localization Error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cumulative Probability Sensor Nodes=800 Sensor Nodes=600 Sensor Nodes=400 Sensor Nodes=200 Figure 3.9: ECDF ALE for various number of sensor nodes Furthermore, the Fuzzy-ELM performances are compared to the original DVHOP-3D and PERLA algorithms. Figure 3.10 shows that Fuzzy-ELM has a lower ALE compared to PERLA and DV-Hop-3D for the same number of nodes. As the number of nodes increases, the ALE progressively decreases for all three algorithms evaluated. The rate of ALE decrease is significant for Fuzzy-ELM for 3.4. Geolocalization via Lightweight Fuzzy-ELM 52 cases from 200 to 400 nodes. In fact, the rate of ALE decrease is higher when the node density is low, and the rate of ALE decrease is lower when the node density is high. As shown in Figure 3.10, the ALE for Fuzzy-ELM falls by around 35% compared to DV-Hop-3D, while compared to PERLA falls by 15%. Whatever the total number of nodes, the results obtained by Fuzzy-ELM algorithm are better than those of PERLA and DV-Hop-3. This simulation results demonstrate the effectiveness of the Fuzzy-ELM localization algorithm in isotropic cases for WSN localization problem, where outperforms the PERLA and DV-Hop-3 algorithms. Figure 3.10: Average localization error versus the total number of sensor nodes 3.4.4 Effect of communication range Our analysis focuses on how different communication ranges affect the average location error (ALE). The communication range of the sensor nodes varies between 25 and 50 m. The total number of sensor nodes is maintained at 400, with 15% of anchor node positions. The communication range variation in Figure3.11 is evaluated against the DV-Hop-3D and PERLA algorithms to examine localization accuracy. For all values in the communication range, Fuzzy-ELM has a lesser ALE compared to DV-Hop-3D and PERLA. In the 25–35 m communication range, the decrease in ALE is more pronounced than in the 35–50 m range. The overall ALE of Fuzzy-ELM decreases by approximately 75% when compared to the DV-Hop-3D and by 50% when compared to PERLA algorithm for a communication range of 50 m. On the basis of these results, we can confirm that our algorithm outperforms the compared algorithms. 3.4. Geolocalization via Lightweight Fuzzy-ELM 53 Figure 3.11: Average localization error versus the communication range of sensor nodes 3.4.5 Robustness analysis for irregular connectivity In the practical application of WSNs, obstacles and electromagnetic noise seriously affect geolocation results, as a result of the non-stability of RSSI signals and the irregularity of their propagation. In this case, the coverage area of the sensor nodes will take the form of an irregular ellipse instead of a perfect circle. In the case of “Range-Free” WSNs, the impact of the irregularity of the communication range on the routing protocol can affect the minimum number of hops that is exploited by the localization system. To characterize the irregularity of RF signal propagation, an irregularity coefficient has been introduced by the scientific community. In fact, the irregularity characterization coefficient (DOI) characterizes the maximum variation in radio range due to changes in the angle of propagation in different directions of the radio antenna. The model of RF signal propagation and connectivity can be described by the following mathematical model [72], where the probability that two sensor nodes can communicate with each other is controlled by the distance between them ( d ) and the ideal communication range R . In this model, the probability of communication between the two sensor nodes is described by the following model: P(d) =                      1,d r<1−DOI, 1 2DOI d r−1+1 2,1−DOI ≤d r≤1+DOI, 0,d r>1+DOI. (3.17) According to this connectivity model and as shown in figure 5.13, the communication range R between two sensor nodes varies according to the value of the DOI. If the DOI is zero, the communication range has the shape of a perfect circle, 3.4. Geolocalization via Lightweight Fuzzy-ELM 54 and as the value of the irregularity coefficient increases, the irregularity of the transmission range is increasingly disturbed (Figure 3.12). Depending on the orientation angle of the RF antenna and climatic conditions, this irregularity is affected, as well as the minimum number of hops between sensor nodes. Taking into account to the DOI coefficient, we implement the Fuzzy-ELM localization and the DOI Model for localization process in Range-free WSN. We assume that the communication range for DOI equal to zero is R=40m and the DOI varies between [0, 0.045]. In these simulation cases, 360 unknown nodes are deployed in a 3-D area delimited by 100 × 100 × 100 m3 with average communication range R=40m and number of anchor nodes equal to 40. 0 10 20 30 40 50 60 70 80 90 100 Meters 10 20 30 40 50 60 70 80 90 Meters DOI=0.01 DOI=0 (a) DOI=0.01 0 10 20 30 40 50 60 70 80 90 100 Meters 10 20 30 40 50 60 70 80 90 Meters DOI=0.045 DOI=0 (b) DOI=0.045 Figure 3.12: DOI irregularity Figure 3.13 illustrates the boxplot of the statistical results of fifty-one times random deployment scenarios for each DOI value. Boxplot shows the mean, max and min of the ALE of localization obtained via the Fuzzy-ELM algorithm. It can be seen that as DOI increases, the connectivity of the wireless sensor network is disrupted and the number of hops between anchors and unknown nodes is affected, leading to a deterioration in localization accuracy. For example, for 40 anchor nodes, if the DOI is zero, the localization error is close to 0.15×R, but if the DOI=0.045, the localization error is close to 0.3×R, because the value of the number of hops is affected and increases localization errors. 3.5. Conclusion 55 0.0 0.010 0.015 0.020 0.025 0.030 0.035 0.040 0.045 DOI 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 Average Localization Error Figure 3.13: Average localization error versus the communication range of sensor nodes Table 3.4 resumes the statistical results of the ALE (mean, Std, Max and Min) for different values of DOI. DOI ALE 0.0 0.010 0.015 0.020 0.025 0.030 0.035 0.040 0.045 mean 0.1735 0.1849 0.2030 0.2424 0.2792 0.2711 0.2760 0.2757 0.3043 max 0.4695 0.5389 0.7224 0.6746 0.7642 0.7315 0.7309 0.6389 0.7289 min 0.0205 0.0217 0.0317 0.0142 0.0414 0.0327 0.0331 0.0451 0.0224 std 0.0745 0.0944 0.1188 0.1152 0.1348 0.1384 0.1339 0.1283 0.1443 Table 3.4: ALE results for different DOI values. 3.5 Conclusion In this chapter, a fuzzy-ELM algorithm has been proposed to improve the accuracy of node localization in WSNs in isotropic cases. The proposed algorithm is based on the range-free technique. Fuzzy-ELM represents a new way of attacking the problem of localization in WSNs. In summary, combining fuzzy logic with extreme learning machine could be a profitable approach that addresses the challenges of vulnerability and complexity in various applications. The performance of the proposed algorithm is well demonstrated by the simulation results in comparison with other algorithms. Thanks to the fuzzification layer for position estimation, the experimental results demonstrate that the Fuzzy-ELM localization algorithm minimizes the average node localization error and achieves higher localization accuracy. These advantages make Fuzzy-ELM a very promising algorithm for dealing with the WSN localization problem in real time. The FPGA implementation of the proposed algorithm will be addressed in the next work. CHAPTER FOUR MOBILE BOUNDING-BOX FOR LOCALIZATION IN WSN 4.1 Introduction The wireless sensor network is composed essentially of sensors capable to collect and transmit data autonomously. The sensors can be deployed randomly in a defined area to detect a phenomenon. The applications of this particular ad-hoc network, Wireless sensor network, are invading many fields such as the IoT, IIOT, e-health, e-logistics, e-agriculture, etc. Some of these applications have potential, such as surveillance of natural parks, surveillance of nuclear power plants, forest fire detection, etc. The main purpose of almost all the applications in WSN is to determine when and where a marked event occurs which is considered an identifier of that occurrence. Some of the nodes constituting a WSN environment are considered anchors that have a stable source of location information via the Global Positioning System. For computational processing scenarios, we consider the advancement of power management strategies for better solutions in particular for low-power WSN using an alternative method than using several GPS devices for a low cost and computational effort. In this work, our main goal is to find an efficient algorithm to achieve highly accurate data and higher localizable nodes with lower computational effort, memory size and energy. We are concerned with the strategy of a mobile anchor node to reduce hardware costs. Our proposal is based on the bounding box approach to identify the possible location area of the unknown node. Also, the proposed approach uses an infinite impulse response filter which is a Kalman filter that approximates the states of a dynamic system from a set of noisy or incomplete information. For wireless sensor 56 4.2. Large-scale WSNs localization algorithms 57 network localization problem, the KF algorithm iteratively refines the positions determined by the bounding box algorithm (BBX) in order to improve the accuracy of the undetermined node positions. Actually, KF uses the dynamic model of the geolocation system, modeled by state equations and onboard sensor measurements, to obtain an estimate of the state vector that minimizes the covariance error. In effect, an important interest oriented to mobility that has a major role in such networks. The combination of mobile systems and WSNs creates what we will be working on, the Mobile Wireless Sensor Networks. In a large-scale network, the cost value is important for the application’s deployment. MWSNs help to solve this problem by decreasing the number of anchors which costs more than normal nodes. Therefore, every localization process needs an anchor node to provide information and references to unknown nodes. Consequently, using the range-free method, specifically, the connectivity process of all nodes estimates their inter-distance via the calculation of the lowest number of hops. Thus, MWSN is capable of deploying a single anchor in the network and determining the positions of unknown nodes based on information about the number of hops and the path of the anchor. The mobile anchor node trajectory could be random or has a predetermined path planning. The purpose is to calculate the connectivity information and then try to compute the positions of unknown nodes using an iterative bounding box. Besides, using a highly utilized filter, Kalman Filter, to refine the localization process and improve localization accuracy. The rest of the paper is organized as follows: After the introduction in section 4.2 Large-scale WSNs localization algorithms Wireless Sensor Networks have become increasingly prevalent in the Internet of Things and Industrial Internet of Things due to their ability to exploit the capabilities of 5G networks. Many research efforts have been focused on addressing various challenges related to WSNs, including location tracking, routing, task allocation, spectral optimization, energy optimization, and more. To improve the performance of these networks, researchers have utilized a range of techniques including machine learning and neural networks [76],[77],[78]. Various types of fields deployed in WSNs require localization techniques to be used in different applications. Some applications such as healthcare treatment, environmental monitoring, and agricultural management require the deployment of nodes in three-dimensional (3D) irregularly shaped fields. The localization of sensor nodes in different types of fields is a challenge [79]. In 3D fields, network connectivity can be more diverse compared to the 2D scenarios and WSN boundary conditions become more important [80]. Range-free localization algorithms are most appro- 4.2. Large-scale WSNs localization algorithms 58 priate for large-scale WSNs since they do not require any additional hardware on the sensor nodes to be used to measure the radio signal characteristics. The most well-known of these algorithms are techniques based on hops [81]. These types of algorithms use only the connectivity between nodes to determine the minimum number of hops and subsequently to estimate the distance [82]. Hop-based techniques use the length of hops between sensor nodes and reference nodes for the estimation of the location. Also, mobile localization techniques do not need additional hardware and estimate the node site area sequentially. The localization with a mobile beacon in WSN was proposed in various research works in literature as a solution for sensor nodes to compute their positions in real time. In fact, the unknown nodes receive three or more signals from the same beacon node from different locations and calculate their locations by applying a trilateration or multilateration algorithm [83]. In this way, we are defining some of the popular range-free methods employed for localization for WSN with the presence of mobility. We can mention the most well-known method Monte Carlo Localization (MCL) algorithm proposed by Lingxuan Hu et al. [84]. Using one mobile anchor [85], the authors presented the Mobile Anchor Centroid Localization (MACL) algorithm for localization in WSNs. The described algorithm is based on radio frequency connectivity and the centroid method for preliminary localization via the single mobile anchor. The considered path planning for the localization process is spiral trajectory. The former algorithm produced acceptable results for some WSN deployment scenarios, but due to the filtering operation, integrated into the algorithm, it is possible to reject the effect of some anchor positions to improve the localization accuracy. In [86], the Mobile Anchor Assisted Localization Algorithm Based on Regular Hexagon (MAALRH) is presented. This method is designed based on the regular hexagon in 2D-WSNs which maximizes the ratio of localized nodes and minimizes the average localization errors. Moreover, for overcoming the irregularity of the communication range and the co-linearity problem, a boundary compensation method (BCM) is described. Furthermore, different path planning for a mobile anchor is presented like HILBERT, CIRCLE and S-CURVE. The research work in [87] presents a mobile anchor node for localization using the trilateration method and path planning choice with two objective functions such as the reduction of localization errors and minimizing the energy consumption for WSNs. Moreover, the inter-distance between the mobile anchor and the unknown nodes is estimated with the RSSI indicator. In addition, the localization errors are analyzed for five trajectories for mobile anchors like SPIRAL, SCAN, DOUBLE SCAN and HILBERT path. The authors of the paper[88] proposed a collaborative localization algorithm with bounding restrictions for wireless mobile networks. The proposed technique implies trajectory prediction using Newton’s polynomial 4.3. The basics of Kalman-Filter 59 interpolation and bounding box. Earlier, most localization techniques were based on numerical computation of many network parameters such as transmitter or receiver power, network connectivity information, transmission range, transmission and reception time, etc. These parameters have many restrictions, such as environmental conditions and complications related to the environment. The Monte Carlo Boxed (MCB) [89] localization method has been proposed by Baggio Aline and Koen Langendoen where the samples are defined by constructing a box on the basis of MCL. This algorithm accelerates the design of position samples, which improves localization accuracy. Many improvements of MCB are presented in the literature, such as RSSI-MCB[90], IMCB [91] and MCB-PSO [92]. Specifically, in range-based cases, many localization techniques in the literature have concentrated on minimizing localization errors during the localization operation. In [93], the uncentered Kalman filter (UKF) and particle filter (PF) were presented based on the Kalman filter (KF) for localization. The published localization technique can be used for any kind of wireless sensor network problem and in particular for Autonomous robot localization. 4.3 The basics of Kalman-Filter The Kalman filter is a method to estimate parameters of a system that evolves in time from noisy measurements. This filter can be found in many fields related to signal processing, radar, image processing, etc. An example of the use of this filter could be the determination of the position and speed of a vehicle from GPS data provided by several satellites. The power of this filter is its capability to fix errors and forecast parameters, both from the sensors and from the actual model. In fact, in order to implement a Kalman filter, the system of which we want to estimate its parameters must be modeled in a linear way. Variants of the Kalman filter exist to take into account non-linear models. In a conventional approach for estimation, such as the least squares method, a single error in the modeling of the system necessarily leads to an error in the estimation. The strength of the Kalman filter is that it integrates an inaccuracy element on the actual model, making it possible to obtain correct estimations in spite of modeling errors (as long as these errors remain reasonable). Another strength of the algorithm is its capacity to determine the average error of its estimate. Indeed, the Kalman tool provides a matrix of the estimated parameters, as well as a covariance matrix of the error. This provides information about the accuracy of the estimate, which might be valuable in many different purposes. Another advantage of the Kalman filter is that the convergence of this error is guaranteed. The Kalman filter process can be divided into two steps. The first step of prediction is according to the model of 4.4. The proposed localization algorithm: BBLA-KF 60 the system. For this, the Kalman filter uses the previous estimation of values for the system’s parameters and errors and predicts the new parameters and errors based on the system model. The second step will update this prediction with the new measurements. These measurements will allow obtaining an estimate of the parameters and the error from the prediction model. The whole process is illustrated in figure 4.1. Figure 4.1: The Kalman-filter algorithm If the model contains errors, this stage of the update will make it possible to correct them. 4.4 The proposed localization algorithm: BBLA-KF Kalman filter is generally widely used with range-based methods for filtering the Received Signal Strength Indication ‘RSSI’ signal to properly estimate distances. Indeed, in our case, it is important to carefully consider features of the hop-based methods such as network coverage, node mobility and communication range, in order to ensure that the range-free mobile anchor sensor network is able to effectively cover and monitor the large scale domain, as well as achieve accurate nodes location in different scenarios. In fact, we propose a novel localization strategy for wireless sensor networks with a mobile anchor node, utilizing a combination of a Kalman filter and bounding box localization algorithm (BBLA-KF). Our approach utilizes connectivity measurement and the nonlinear bounding box techniques to identify the bounded possible location zone for the unknown nodes, and then applies the kalman filter to minimize uncertainty in the connectivity process and achieve high levels of accuracy in determining the bounded-boxes. 4.5. BBLA-KF simulation results 67 4.5.1 Experimental analysis of positioning We use MATLAB to build our sensing area, in order to follow up on the point that all nodes are able to be located. We distribute 340 unknown nodes and 60 anchor nodes randomly and uniformly in the deployment zone. Figure 4.6: 3D-WSN Nodes distribution Nodes distribution is shown in figure 4.6, where yellow stars represent anchor node positions, and red (*) represent unknown nodes. We examine our method in the environment shown in Fig.4.6. Then, we get the result of MATLAB simulation shown in figure 4.7, where (*) is the true node location and (o) the estimated location. Figure 4.7: Localization results of our proposed algorithm From the graph, we can see that the BBLA-KF method can locate the unknown nodes very well, and give a higher precision. Figure 4.8 is the result obtained after conducting a number of tests for evaluating the convergence of the BBLA-KF 4.5. BBLA-KF simulation results 68 Figure 4.9: Average localization error of unknown nodes over the number of iterations algorithm towards the real positions. From the figure, we can see the efficiency of our proposed algorithm to converge. Figure 4.8: Convergence of BBLA-KF algorithm towards the real positions Figure 4.9give the convergence of ALE over the number of iterations. This behavior may be due to the use of the Kalman filter in position adjustment, which utilizes a set of observed measurements through time, including statistical noise and other imprecisions, and generates estimates of unknown variables that tend to be more precise. Figure 4.10 shows the location error of all unknown nodes. It can be seen that the biggest error does not exceed 7m equivalent to 40% of R, communication range. The error range is acceptable for mobile WSN Localization. 4.5. BBLA-KF simulation results 69 Table 4.1: Localization error results of hundred trials Localization erorr Localization algorithms Max Min Mean BBLA-KF 0.25 0.10 0.23 I3D-DVLAIN 0.37 0.04 0.25 PERLA 0.38 0.27 0.35 OIDV-Hop 0.45 0.35 0.39 DV-HOP-3D 0.42 0.25 0.37 Figure 4.10: The error given by each unknown node 4.5.2 Comparative analysis with other localization techniques Following, we present some simulation results that compare the localization accuracy performance of our proposed algorithm BBLA-KF versus RRBL, OIDVHop, PERLA, I3D-DVLAIN and DV-Hop-3D. We perform simulations to compare the performance in the same scenario set in section sec4:The Basics of Kalman-Filter The I3DDVLAIN, OIDV-Hop, PERLA, and DV-Hop-3D algorithms were run for one hundred aleatory displays and the percentage of localization error for all trials is shown in Table 4.1. Compared to other existing algorithms, the results of our algorithm are better. When we look at the average value at the end of 100 trials, we find that BBLA-KF has an average of 23% while I3D-DVLAIN has an average LE of about 25%, DV-Hop has about 37%, OIDV-Hop has about 39%, and PERLA has about 35%. The resulting localization error of BBLA-KF is the lowest among the five techniques compared. This shows that our newly proposed algorithm is much more accurate than the other algorithms. We may justify this with the fact that our proposed algorithm is built on both the bounding box method and the Kalman filter to match the estimated position. Therefore, BBLA-KF performs better in terms of localization accuracy. 4.5. BBLA-KF simulation results 70 Table 4.2: Simulation parameters of nodes number variation Parameters values Sensing field 100m×100m×100m Communication range 18m Node amount 100,200,300,400,500,600,700,800 Anchor ratio 15 % 4.5.3 Influence of the variation of the number of nodes The performance of the presented algorithms is evaluated for various node densities in 3D fields. The number of nodes (anchor nodes and unknown nodes) is increased from 100 to 800 and the ratio of reference nodes is kept at 15% with a communication range of 18 m. The simulation parameters are presented in Table 4.2. The simulation results of the influence of the number of nodes are illustrated in Figure 4.11. Figure 4.11: Average localization error versus the total number of sensor nodes We notice that the ALE for BBLA-KF is smaller than that of I3D-DVLAIN, OIDV-Hop, PERLA and DV-Hop-3D for the same value of the total number of nodes. The ALE for DV-Hop-3D, OIDV-Hop, and PERLA decreases slowly as compared to BBLA-KF. The grade in ALE is very significant from 100 to 300 nodes for BBLA-KF, which means that the rate of ALE improvement was elevated at lower node densities and the variation in ALE was decreased at greater node densities. Also, there is a fall of around 40% in ALE for BBLA-KF as compared to DV-Hop-3D and a fall of 20% with respect to I3D-DVLAIN, as seen in Figure 4.11. Figure 4.12 gives the cumulative distribution functions (CDFs) of the average localization error for different WSN densities. For example, in this figure, we notice that for 100 nodes, 14% of the unknown nodes are localized with an error 4.5. BBLA-KF simulation results 71 less than 0.25 of the communication range "R", while for 200 nodes, only 42% of the unknown nodes are localized with an error inferior to 0.25R. However, for 400 nodes the localization error is 91% and for 800 nodes, all the unknown nodes are localized with an error less than the previous error range. Next, table 4.3 resume the CDFs of localization error for different number of nodes. Table 4.3: Localization error CDF’s for different number of nodes Localization erorr Number of nodes 0.25R 0.5R R 800 100% 100% 100% 400 91% 100% 100% 200 42% 78% 100% 100 14% 57% 95% We can interpret from the table 4.3 that for the case of 100 nodes, there are some nodes non-localizable which is explained by the fact that the localization error is greater than the communication range. We also find that as the number of nodes decreases, the CDFs tend to decrease. This may be due to the growth in network connectivity, which is intended to give an accurate estimate of the distance between nodes in large-scale WSNs. In conclusion, we can say that increasing the density of nodes improves the performance of the algorithm by improving the accuracy. Figure 4.12: Cumulative distribution function (CDF) plots of different number of sensor nodes 4.5.4 Influence of anchor node positions ratio The communication range of the sensor nodes has been set to 18 m with an error ranging from 0 to 10%. The total number of sensors in the network is set 4.5. BBLA-KF simulation results 72 to 400. Therefore, we vary the anchor node positions ratio from 5 to 35% from the total number of nodes and analyze the result with respect to its localization error. Figure 4.11 shows a comparison of the localization error of the algorithms I3D-DVLAIN, DV-Hop-3D, OIDV-Hop, PERLA and BBLA-KF. It can be clearly seen that as the ratio of anchor node positions increases, the localization error of the different algorithms decreases because the unknown nodes receive more accurate information due to the multiple anchor node positions. Similarly, for various values of anchor node position ratio, the localization error of BBLA-KF is always lower than that of other algorithms due to the position correction applied by the Kalman filter for unknown nodes. Figure 4.13: Localization error versus percentage of anchor nodes 4.5.5 Effect of communication range Next, we analyze the effect of varying communication range on the average location error (ALE). The communication range of sensor nodes varies from 15m to 45m. The total number of nodes has been kept to 400 which contains 15% of anchor node positions. Figure 4.14 presents the effect of the variation in the communication range on the localization accuracy by comparing with I3D- DVLAIN, DV-Hop-3D, OIDV-Hop and PERLA algorithms. The ALE of BBLA-KF is lower than that of I3D-DVLAIN, DV-Hop-3D, OIDV-Hop and PERLA for each value in the communication range. The decrease in ALE is more severe in the 15- 25m communication range than in the 35-45m range, where the ALE value almost normalizes. There is an approximately 40% decrease in overall ALE for BBLA-KF compared to the I3D-DVLAIN algorithm. The effect of varying communication range on the ALE for our algorithm is also smaller compared to other algorithms. Based on these results, we are able to confirm that our algorithm performs better than other algorithms. 4.5. BBLA-KF simulation results 73 Table 4.4: ALE results for isotropic and anisotropic 3D fields 3D-WSN fields Average Localization Error Isotropy 0.21 Anisotropy 0.26 Figure 4.14: Localization error versus percentage of communication range 4.5.6 Effect of anisotropy BBLA-KF is evaluated in 3D fields. The impact of anisotropy is evaluated by studying the localization results in anisotropic fields with isotropic fields. Thus, The results of localization of 3D cubic field and 3D C-shaped field are compared for our proposed algorithm. Using the MATLAB simulation experiment, we obtain the experimental results in figure 4.15, figure4.16, where the red lines represent the location error, the blue circles are the reel positions of the unknown nodes and the red circles are the estimated positions of the unknown nodes by our algorithm. Figure 4.16, the C-shaped anisotropic field and figure4.15 the isotropic field give a good ALE which means a good localization accuracy. The BBLA-KF shows good results in both isotropy and anisotropic fields, as mentioned in table 4.4, which proves the efficiency of our algorithm. 4.6. Conclusion 74 Figure 4.15: Location results for unknown nodes in 3D cubic field Figure 4.16: Location results for unknown nodes in 3D C-shaped field 4.6 Conclusion The precision of localization algorithms is a problem, especially in irregular areas. The majority of existing algorithms in the literature have been developed for locating nodes in regular fields, whereas the real-world applications of WSNs are in irregular fields. In order to help overcome this gap, we proposed a localization algorithm called BBLA-KF. This hop-based algorithm uses properties of the Bounding-Box and Kalman-Filter methods to enhance localization accuracy in a variety of field types. The results of this algorithm show better localization accuracy in various 3D fields of different shapes compared to its counterparts recently published. In future work, we are planning to implement this algorithm on an actual WSN to solve the localization problem. Also, We are planning to 4.6. Conclusion 75 design and develop our model on a configurable Field Programmable Gate Array (FPGA). CHAPTER FIVE REGULARIZED BBLA-KF LOCALIZATION ALGORITHM AND HARDWARE IMPLEMENTATION 5.1 Introduction In the emergent wireless sensor network, accurate localization of sensor nodes is a crucial challenge for many potential applications, such as smart cities, intelligent navigation, process supervision and control, IoTs devices and military tracking applications[98][99][100]. For example, in military application Figure 5.1, it is often necessary to determine the position of a dozen or even hundreds of soldiers in a war zone for rescue operations, where each soldier is equipped with wireless LORA or Zigbee technology sensors to communicate with a drone in order to locate the soldiers or gather information about the environment[101]. In the literature, a limited number of research works have been carried out to implement localization algorithms on a dedicated hardware architecture (see Related Works section). In this work, we therefore present a complete procedure for implementing and evaluating a Regularized Bounding-Box within the Kalman Filter algorithm (RBBLA-KF) for localizing unknown nodes in an area using an FPGA-enhanced mobile UAV that employs edge computing to perform real-time processing and analysis of sensory data without relying on remote servers or cloud-based computing resources, which brings computational tasks and data storage closer to the sources of data (Figure 5.2), unlike traditional cloud computing, where data is collected and processed centrally. This can reduce latency since data does not have to be transmitted over long distances to a remote server[102][103][104]. Also, it can improve performance, since data can be processed and analyzed closer to the source. Moreover, it can increase the reliability 76 5.3. Regularized BBLA-KF algorithm for localization in WSN 83 Xmes =           αx1−αx0 0 0 0 0 0 αy1−αy0 0 0 0 0 0 αz1−αz            ·                             x+ x− y+ y− z+ z−                             (5.3) where αx,αy,αz represent the regularization factors to be adjusted and depend on the regularity of the deployment zone (isotropic or anisotropic case) and the trajectory of the mobile anchor. In our simulation, these parameters are set by a trial-and-error procedure. The bounding box is defined by a set of coordinates that enclose an object and is typically represented as a set of six geographical coordinates for their three-dimensional field: the x− is the left side, x+ is the right side, y− is the bottom side, y+ is the top side, z− is the back side and z+ is the front side of the bounded box, which are then processed to determine the unknown nodes’ positions. One of the main advantages of using bounding boxes for object localization is that they are simple and efficient to compute [126][42]. The bounding box is also highly customizable, quickly and easily generated and can be used in a wide range of applications. In addition to the regularized bounding box, we also used the commonly used method Kalman filter, for localizing and identifying sensor nodes. The Kalman filter uses a mathematical model of the system to combine the available measurements with our prior knowledge of the system’s state, given by the equation 4.7, to produce an estimate of the current state. In this scenario, we consider the position of the unknown node as fixed and unaltered during the localization process, so the value of wk is null. For this system, the measurement equation is mentioned in equation 4.8. In the case of UAV, the Kalman filter uses the UAV’s GPS measurements and the number of hops received from each unknown node to estimate the positions of the unknown nodes. The GPS measurements provide information about the UAV’s own position, while the number of hops provides information about the distance between the UAV and the unknown nodes. By combining these two types of information, our algorithm is able to produce accurate estimates of the unknown nodes’ positions. They can also be tailored to the specific needs of a given application. Together, the Kalman filter and regularized bounding box method provide a robust and accurate approach for localization, allowing the UAV to navigate and perform tasks in unknown environments. The Kalman filter provides reliable estimates of the unknown nodes’ positions, while the regularized bounding box method provides the bounding box limits for each unknown node. This combination of techniques allows the UAV to accurately 5.4. Simulation results 84 locate the surrounding nodes, enabling it to navigate and perform tasks with confidence. Figure 5.6 gives the algorithm structure of the proposed RBBLA-KF. The box on the left first describes the flooding phase, which consists of finding the number of hops separating the mobile anchor and the unknown nodes. Next, the regularized mobile bounding box algorithm is applied via the equations (5.1) (5.2) (5.3). The box on the right presents the Kalman filter algorithm for the refinement process to improve the estimated position given by the regularized mobile bounding box. Figure 5.6: The proposed RBBLA-KF algorithm to be implemented on FPGA 5.4 Simulation results The performance of our RBBLA-KF algorithm is being evaluated through studies on the properties of a 3D wireless sensor network. The average localization error (ALE) is presented to analyze the impact of varying the regularization factors on the results of sensor node localization. ALE =PN 1p(xk,mes −xk)2+(yk,mes −yk)2+ (zk,mes −zk)2 R×N(5.4) 5.4.1 Regularization factors effects In order to improve the accuracy of sensor node localization in our case study, we studied the effects of varying regularization factors ( αx , αy , αz ) on the average localization error. The results presented in this section compare the performance of the proposed method under the same conditions of the 3D WSN. We will conduct experiments with 400 sensor nodes randomly distributed across a 100m 5.4. Simulation results 85 × 100m × 100m detection zone without obstacles. In order to ensure the drone’s connectivity with the WSN, we set the communication range at 30m, respecting the condition given by the equation 5.5. R≥ZUAV max (5.5) where ZUAV max represents the height of the drone. Table 5.1 shows the parameters of the 3D wireless sensor network being simulated, such as deployment area, communication range, number of unknown nodes and UAV trajectory planning. Table 5.1: Simulation parameters Parameters values Deployement area 100m×100m×100m Communication range 30 m Node amount 400 UAV trajectory RWP with ZUAV max ≤30m We used MATLAB-2016 software to determine performance measurement simulations. The path-planning of the UAV is a random waypoint trajectory (RWP). αz0 0.25 0.5 0.75 1 ALE 1.051 1.780 0.6209 0.3066 0.1889 Table 5.2: αzvariation (αy= 0.5; αx=0.5) αy0 0.25 0.5 0.75 1 ALE 0.3869 0.3064 0.1949 0.1006 0.1358 Table 5.3: αyvariation (αz= 1; αx= 0.5) αx0 0.25 0.5 0.75 1 ALE 0.2376 0.1570 0.1111 0.0979 0.1532 Table 5.4: αxvariation (αz= 1; αy= 0.75) We performed various simulations, as mentioned in Tables 5.2, 5.3, and 5.4, to determine the best balance between accuracy and RBBLA-KF regularization factors. The optimal regularization factor for the z-axis equal to one is explainable by the unique spatial constraints imposed by the UAV’s positioning relative to 5.4. Simulation results 86 the sensors. In this context, the regularization along the z-axis eliminates the upper part of the bounding box. Given that the UAV is situated above the sensors during tracking, the upper side of the bounding box is inherently limited by the presence of the UAV itself. Consequently, setting the regularization factor to 1 effectively eliminates the need for an upper limitation in this dimension, aligning the bounding box more precisely with the physical constraints of the UAV’s environment and optimizing the localization algorithm for improved performance. The sample results illustrated in Figure 5.7, 5.8 and 5.9 were obtained for the ideal combination of regularization factors ( αz = 1, αy = 0 . 75, αx = 0 . 75) for evaluating the RBBLA-KF algorithm in different scenarios. The algorithm is evaluated in various deployment zone types: 3D Cubic field, 3D C-shaped field, 3D A-shaped field. These fields are chosen as a combination of isotropic and anisotropic 3D areas. In these figures, the green lines represent the UAV’s trajectory, and the green dots represent the UAV’s positions during the localization process. The blue circles ’o’ represent the estimated positions of the unknown nodes given by RBBLA-KF and the orange stars ’*’ represent the real positions of the sensor nodes. The red interrupted lines represent the estimated errors. 0 100 80 60 Y-axis 40 20 20 Z-axis 100 90 X-axis 80 70 60 50 40 030 20 10 0 40 UAV trajectory estimated positions unknown nodes Figure 5.7: Localization results of our proposed algorithm in 3D Cubic field 5.4. Simulation results 87 Figure 5.8: Localization results of our proposed algorithm in 3D C-shaped field Figure 5.9: Localization results of our proposed algorithm in 3D A-shaped field 5.4.2 Qualitative performances In this section, in order to assess the performance of the RBBLA-KF algorithm versus the BBLA-KF algorithm, as well as to understand the reliability of the statistical results obtained and the robustness of the algorithm, fifty-one independent runs were carried out for various WSN configurations.Figure 5.10 depicts the boxplots simulation results that show the impact of the number of nodes with a communication range equal to 30m and the regularization factors fixed to ( αx = 0 . 75, αy = 0 . 75, αz = 1). The ALE for RBBLA-KF mentioned in Figure 5.10b is inferior to that of BBLA-KF illustrated in Figure 5.10a for the same total number of 5.4. Simulation results 88 nodes which confirms that the impact of the regularization procedure can improve localization errors with different scenarios regarding the number of sensor nodes. Figure 5.11 shows the empirical cumulative distribution function of ALE. We note that the performance of RBBLA-KF surpasses the results of BBLA-KF. Indeed, Figure 5.11b gives the cumulative probability for RBBLA-KF’s ALE where 90% of unknown sensor nodes are located with an average localization error lower than 0.26 × R meters, whereas Figure 5.11a for BBLA-KF’s cumulative probabilty gives an accuracy equal to 0.29 × R meters for the same percentage of unknown nodes to be localized. We also find that as the number of nodes increase, the ECDFs tend to decrease. This may be due to the growth in network connectivity, which is intended to give an accurate estimate of the distance between nodes in large-scale WSNs. In conclusion, we can say that increasing the density of nodes improves the performance of the algorithm by improving the accuracy. Table 5.5 shows the statistical results in the form of mean, standard deviation (Std), maximum and minimum of the ALE results for fifty-one independent simulation runs. In fact, this table proves the superiority of RBBLA-KF over BBLA-KF. 100 200 300 400 500 600 700 800 Number of nodes 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Average Localization Error (a) BBLA-KF’s Boxplot 100 200 300 400 500 600 700 800 Number of nodes 0.1 0.15 0.2 0.25 0.3 0.35 0.4 Average Localization Error (b) RBBLA-KF’s Boxplot Figure 5.10: Average localization error versus the total number of sensor nodes 5.4. Simulation results 89 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34 0.36 Average Localization Error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cumulative Probability 100-Nodes 200-Nodes 300-Nodes 400-Nodes 500-Nodes 600-Nodes 700-Nodes 800-Nodes (a) BBLA-KF 0.18 0.2 0.22 0.24 0.26 0.28 0.3 0.32 0.34 0.36 Average Localization Error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cumulative Probability 100-Nodes 200-Nodes 300-Nodes 400-Nodes 500-Nodes 600-Nodes 700-Nodes 800-Nodes (b) RBBLA-KF Figure 5.11: ECDF ALE for various number of sensor nodes Table 5.5: Statistical results for different number of nodes Number of Nodes 100 200 300 400 500 600 700 800 BBLA-KF Mean 0.2657 0.2596 0.2600 0.2568 0.2613 0.2565 0.2554 0.2572 Std 0.0247 0.0172 0.0178 0.0202 0.0201 0.0170 0.0155 0.0171 Min 0.2208 0.2263 0.2264 0.2195 0.2230 0.2258 0.2270 0.2242 Max 0.3494 0.3097 0.3139 0.3176 0.3193 0.3016 0.2879 0.2975 RBBLA-KF Mean 0.2309 0.2280 0.2290 0.2268 0.2298 0.2207 0.2277 0.2246 Std 0.0295 0.0269 0.0223 0.0169 0.0188 0.0171 0.0196 0.0135 Min 0.1826 0.1822 0.1937 0.1938 0.1973 0.1907 0.1884 0.1962 Max 0.3033 0.3001 0.3039 0.2756 0.2868 0.2706 0.3049 0.2654 Afterward, we examine the impact of varying communication range on the average location error (ALE) for 400 sensor nodes deployed in the 3D WSN. The communication range of sensor nodes extends from 20 to 40 meters. The boxplots given in Figure 5.12 present the effect of the communication range variation on the ALE. The comparison shows that the ALE of RBBLA-KF (Figure 5.12b) is lower than BBLA-KF (Figure 5.12a) for each value in the communication range. It can be clearly seen that as the communication range increases, the localization error of both algorithms decrease because the unknown nodes receive more accurate information from the neighboring nodes due to the large communication range of the sensor nodes. Similarly, for various values of communication range, the localization error of RBBLA-KF is always lower than BBLA-KF due to the position correction applied by the regularized factors. There is an approximately 20% decrease in overall ALE for RBBLA-KF compared to the BBLA-KF algorithm. Figure 5.13 gives the ECDFs of the average localization error while varying the communication range from 20m to 40m. For example, for a range of 20m, for BBLA-KF in Figure 5.13a , we notice that 90% of the unknown nodes are localized with an error less than 0.45 × R meters, while for RBBLA-KF in Figure 5.13b, with 5.4. Simulation results 90 an error inferior only to 0.4 × R meters. And for 40m of communication range, the BBLA-KF localise 90% of the unknown nodes with an error lower than 0.22 × R. However, for RBBLA-KF, the unknown nodes are localized with an error less than 0.17 × R. Next, table 5.6 resumes the statistical results of the ALE (mean, Std, Max and Min) for different values of communication range where the results of BBLAKF are compared to the regularized RBBLA-KF and the resulting localization error is lower than the former algorithm. This shows that our newly proposed algorithm is much more accurate. 20 25 30 35 40 Communication Range 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 Average Localization Error (a) BBLA-KF’s Boxplot 20 25 30 35 40 Communication Range 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 Average Localization Error (b) RBBLA-KF’s Boxplot Figure 5.12: Average localization error versus the Communication Range 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Average Localization Error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cumulative Probability 20 C-Range 25 C-Range 30 C-Range 35 C-Range 40 C-Range (a) BBLA-KF 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5 Average Localization Error 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Cumulative Probability 20 C-Range 25 C-Range 30 C-Range 35 C-Range 40 C-Range (b) RBBLA-KF Figure 5.13: ECDF accuracy for various Communication Range 5.5. FPGA implementation of the RBBLA-KF algorithm 91 Table 5.6: Statistical results for different values of communication range Communication Range (meter) 20 25 30 35 40 BBLA-KF Mean 0.3897 0.3068 0.2600 0.2213 0.1925 Std 0.0284 0.0207 0.0178 0.0176 0.0117 Min 0.3407 0.2672 0.2264 0.1925 0.1684 Max 0.4640 0.3663 0.3139 0.2714 0.2189 RBBLA-KF Mean 0.3379 0.2448 0.2290 0.1898 0.1501 Std 0.0332 0.0192 0.0223 0.0143 0.0112 Min 0.2625 0.1934 0.1937 0.1613 0.1279 Max 0.4272 0.2939 0.3039 0.2277 0.1702 5.5 FPGA implementation of the RBBLA-KF algorithm This work aims to implement a range-free localization algorithm on FPGA via MATLAB/Simulink software using pre-defined Xilinx blocksets and models. The algorithm is designed explicitly for localization tasks in WSN. The hardware implementation of the algorithm is crucial to achieving real-time performance for various localization applications. To accomplish FPGA implementation, we will employ prototyping using the high-level programming environments of Matlab, Xilinx System Generator and Vivado. The process comprises three primary phases: • Input pre-processing blocks: These blocks prepare the input data, in the form of hops, for subsequent localization stages by applying suitable preprocessing techniques. The specific pre-processing units have been described in our previous work [127] and are represented as Simulink blocks. • Algorithm modeling and design using XSG: The localization algorithm will be designed and modeled using Xilinx System Generator blocks. This phase focuses on creating an efficient hardware implementation of the localization algorithm. • Output post-processing blocks: Following the localization process, the output positions will be subjected to post-processing stages for an exploitation phase. The proposed design flow for hardware implementation of the localization algorithm is depicted in Figure 5.14. 5.5. FPGA implementation of the RBBLA-KF algorithm 92 Figure 5.14: The system architecture of the proposed algorithm 5.5.1 Work methodology and tools In this paper, we have explored two distinct implementation phases for the localization algorithm system. The first phase, software implementation, involved a MATLAB-based system, which was employed for testing and validating the proposed methodology. The second phase, hardware implementation, utilized an FPGA-based system to develop a hardware prototype of the system. To begin the process of FPGA implementation, an initial step involves the utilization of the Xilinx System Generator to implement the localization algorithm. This implementation serves as a valuable high-level programming tool, enabling the assessment of algorithm performance in handling complex calculation tasks. By employing a software implementation based on MATLAB, it becomes feasible to construct the hardware implementation while simultaneously evaluating its overall effectiveness. On the other hand, the FPGA-based system that has been chosen in our work is MiniZed™, a single-core Zynq 7Z007S development board. This board targets entry-level Zynq developers with a low-cost prototyping platform, the Zynq xc7z007s-1clg225. Since this board has a small size and the power consumption is very low (5 V/1.5 A), it could equip any mobile engine like a UAV or a mobile robot, and it is also cheaper compared to the other boards. 5.5.2 FPGA implementation Within this section, we elaborate on the procedural details of implementing our localization algorithm on a SoC-FPGA hardware architecture. Our approach commenced with utilizing the Xilinx System Generator for the design and simulation phases via Matlab/Simulink. Then, Vivado is used for synthesizing and implementing the XSG-generated VHDL code and programming the FPGA with 5.6. Conclusion 99 hardware realization of our localization application with the MiniZed Xilinx Zynq development board. Figure 5.21: Real-time implementation of localization algorithm 5.6 Conclusion In summary, our implementation of a localization algorithm on an FPGA board for an edge-computing UAV demonstrates the potential of the embedded regularized mobile bounding box algorithm enhanced by the Kalman filter for localization in large-scale wireless sensor networks. The use of the Kalman filter and the regularized bounding box method allows precise localization of surrounding nodes, enabling the robot to operate efficiently for any WSN range-free scenario. The implementation of our proposed approach on the SoC-FPGA Minized evaluation kit using the Xilinx system generator and Vivado HLS software was successful. The performance of the system was evaluated in terms of resources: LUT, Slice registers, and DSP. Finally, in this article, we demonstrate the advantages of hardware implementation over software implementation in terms of power consumption, execution time and resources used, which is crucial for wireless network applications. CONCLUSIONS AND FUTURE PROSPECTS Conclusions In order to improve the performance of multi-hop localization schemes, a new localization method in static range-free WSNs exploiting the combination of extreme machine learning and fuzzy logic methods has been presented. In addition, a new localization method in range-free mobile WSNs exploiting the bounding box and Klaman filter is presented, along with the process of implementing the proposed approach on FPGA hardware architecture. This chapter summarizes our research and indicates future directions for a possible extension of this work. Thesis contributions This thesis was the subject of three contributions, which can be summarized as follows: • We proposed a fuzzy-ELM learning machine to provide precise geolocation in WSNs where localization is independent of hardware and computational distance. The method for training WSNs combines two techniques within the field of artificial intelligence: the ELM combined with the concept of Fuzzy-Logic (Fuzzy-ELM) is given to improve localization accuracy in WSNs in the Range-Free case. The proposed method is applied in different multihop WSN scenarios. We have focused on the isotropic case and irregular environments. The simulation results show that the proposed Fuzzy-ELM algorithm considerably minimizes the geolocation error compared with other algorithms based on soft computing techniques. • We proposed a new localization algorithm, RBBLA-KF. It focuses on improving our proposed algorithm BBLA-KF. We employ a regularized bounding box to predict the positions of the unknown nodes using regularization factors and we use the Kalman filter algorithm to smooth and improve the 100 Future prospects 101 accuracy of the sensor nodes to be localized. The proposed algorithm is based on a recursive calculation of the position of unknown nodes. Our goal is to use a regularized formulation to compute the average position of specific nodes that don’t know their positions in the network, in order to minimize localization error and achieve better localization accuracy. Simulation results demonstrate the outperformance of our proposed algorithm in comparison with the BBLA-KF algorithm and algorithms proposed in the literature in different deployment scenarios. • The localization algorithm RBBLA-KF was implemented on the FPGA board “Zynq MiniZed 7007s evaluation board” using Xilinx blocks in Simulink, and the generated code was converted into VHDL using Xilinx System Generator. The algorithm was simulated and synthesized using “Vivado” software. In fact, the proposed system was evaluated by comparing the performances achieved through two different implementations: Hardware and Software implementation. Future prospects Real wireless environments are subject to variations in channel quality, interference and node density, which are likely to affect location errors in threedimensional wireless networks (3D-WSNs). A more accurate model for inter-node distance estimation could involve the use of a Rayleigh or Rician distribution for hop quantization, particularly in scenarios where multipath fading can be used in these cases. The use of the uniform distribution with a Rician distribution, which takes into account LOS and NLOS conditions for hop size estimation. This problem requires an analysis of the robustness of the proposed localization process, which needs to be carried out in relation to sensor node mobility and cyber-attacks on sensor nodes or wireless sensor networks embedded in a real operating environment. Also, we aim to investigate how Generative Adversarial Networks (GANs) can generate optimized path planning in WSNs, enhancing data collection, coverage, and energy efficiency. The utilization of GANs for trajectory planning holds great potential to optimize autonomous navigation, considering factors like node density and environmental dynamics, thereby improving node location accuracy in real-world applications. CONCLUSIONES Y PERSPECTIVAS Conclusiones Con el fin de mejorar el rendimiento de los esquemas de localización multisalto, se ha presentado un nuevo método de localización en WSN estáticas sin cobertura que explota la combinación de métodos de aprendizaje automático extremo y de lógica difusa. Además, se presenta un nuevo método de localización en WSN móviles sin alcance que explota la caja delimitadora y el filtro de Klaman, junto con el proceso de implementación del enfoque propuesto en la arquitectura de hardware FPGA. Este capítulo resume nuestra investigación e indica futuras direcciones para una posible extensión de este trabajo. Contribuciones a la tesis Esta tesis ha sido objeto de tres contribuciones, que pueden resumirse como sigue: • Proponemos una máquina de aprendizaje fuzzy-ELM para proporcionar una geolocalización precisa en WSNs donde la localización es independiente del hardware y de la distancia computacional. El método para entrenar WSNs combina dos técnicas dentro del campo de la inteligencia artificial: el ELM combinado con el concepto de Fuzzy-Logic (Fuzzy-ELM) se da para mejorar la precisión de localización en WSNs en el caso Range-Free. El método propuesto se aplica en diferentes escenarios de WSN multisalto. Nos hemos centrado en el caso isotrópico y en entornos irregulares. Los resultados de la simulación muestran que el algoritmo Fuzzy-ELM propuesto minimiza considerablemente el error de geolocalización en comparación con otros algoritmos basados en técnicas de soft computing. • Hemos propuesto un nuevo algoritmo de localización, RBBLA-KF. Se centra en mejorar nuestro algoritmo propuesto BBLA-KF. Empleamos un bounding 102 Perspectivas de futuro 103 box regularizado para predecir las posiciones de los nodos desconocidos utilizando factores de regularización y utilizamos el algoritmo del filtro de Kalman para suavizar y mejorar la precisión de los nodos sensores a localizar. El algoritmo propuesto se basa en un cálculo recursivo de la posición de los nodos desconocidos. Nuestro objetivo es utilizar una formulación regularizada para calcular la posición media de nodos específicos que desconocen su posición en la red, con el fin de minimizar el error de localización y lograr una mayor precisión en la localización. Los resultados de la simulación demuestran el rendimiento superior de nuestro algoritmo propuesto en comparación con el algoritmo BBLA-KF y los algoritmos propuestos en la literatura en diferentes escenarios de despliegue. • El algoritmo de localización RBBLA-KF se implementó en la tarjeta FPGA «Zynq MiniZed 7007s evaluation board» utilizando bloques Xilinx en Simulink, y el código generado se convirtió a VHDL utilizando Xilinx System Generator. El algoritmo se simuló y sintetizó utilizando el software «Vivado». De hecho, el sistema propuesto se evaluó comparando las prestaciones alcanzadas mediante dos implementaciones diferentes: Hardware y Software. Perspectivas de futuro Los entornos inalámbricos reales están sujetos a variaciones en la calidad del canal, las interferencias y la densidad de nodos, que probablemente afecten a los errores de localización en redes inalámbricas tridimensionales (3D-WSN). Un modelo más preciso para la estimación de la distancia entre nodos podría implicar el uso de una distribución Rayleigh o Rician para la cuantificación de los saltos, sobre todo en escenarios en los que en estos casos puede utilizarse el desvanecimiento multitrayecto. El uso de la distribución uniforme con una distribución de Rician, que tiene en cuenta las condiciones LOS y NLOS para la estimación del tamaño de los saltos. Este problema requiere un análisis de la robustez del proceso de localización propuesto, que debe llevarse a cabo en relación con la movilidad de los nodos sensores y los ciberataques a nodos sensores o redes de sensores inalámbricas embebidas en un entorno operativo real. Además, pretendemos investigar cómo las Redes Generativas Adversariales (GANs) pueden generar una planificación optimizada de trayectorias en WSNs, mejorando la recogida de datos, la cobertura y la eficiencia energética. La utilización de GANs para la planificación de trayectorias tiene un gran potencial para optimizar la navegación autónoma, teniendo en cuenta factores como la densidad de nodos y la dinámica del entorno, mejorando así la precisión de la localización de nodos en aplicaciones del mundo real. BIBLIOGRAPHY [1] R. Mani, A. Rios-Navarro, J. - L. Sevillano-Ramos, and N. Liouane, “Improved 3d localization algorithm for large scale wireless sensor networks”, Wireless Networks, pp. 1–16, 2023. [2] R. Mani, A. Rios-Navarro, J. L. S. Ramos, and N. Liouane, “Localizing unknown nodes with an fpga-enhanced edge computing uav in wireless sensor networks: implementation and evaluation”, Pervasive and Mobile Computing, p. 101 961, 2024. [3] R. Mani, J. L. Sevillano, and N. Liouane, “Improved least-square dv-hop algorithm for localization in large scale wireless sensor network”, in 2022 International Conference on Control, Automation and Diagnosis (ICCAD), IEEE, 2022, pp. 1–6. [4] R. Mani, S. Messous, G. Campobello, F. Battaglia, and N. Liouane, “Improved distance vector based kalman filter localization algorithm for wireless sensor network”, in 2023 International Conference on Control, Automation and Diagnosis (ICCAD), IEEE, 2023, pp. 1–6. [5] T. Arampatzis, J. Lygeros, and S. Manesis, “A survey of applications of wireless sensors and wireless sensor networks”, in Proceedings of the 2005 IEEE International Symposium on, Mediterrean Conference on Control and Automation Intelligent Control, 2005., 2005, pp. 719–724. doi: 10.1109/ .2005.1467103. [6] K. Khedo, P. Rajiv, and M. Avinash, “A wireless sensor network air pollution monitoring system”, International Journal of Wireless & Mobile Networks, vol. 2, May 2010. doi:10.5121/ijwmn.2010.2203. [7] N. Vlajic and N. Moniz, “Self-healing wireless sensor networks: results that may surprise”, in 2007 IEEE Globecom Workshops, 2007, pp. 1–6. doi: 10.1109/GLOCOMW.2007.4437830. [8] S. Kumar and D. Shepherd, “Sensit: sensor information technology for the warfighter”, in Proc. 4th Int. Conf. on Information Fusion, 2001, pp. 1–7. [9] A. Redondi, M. Chirico, L. Borsani, M. Cesana, and M. Tagliasacchi, “An integrated system based on wireless sensor networks for patient monitoring, localization and tracking”, Ad Hoc Networks, vol. 11, no. 1, pp. 39–53, 2013. [10] K. S. Adu-Manu, F. Engmann, G. Sarfo-Kantanka, G. E. Baiden, and B. A. Dulemordzi, “Wsn protocols and security challenges for environmental monitoring applications: a survey”, Journal of Sensors, vol. 2022, no. 1, p. 1628537, 2022. 104 Bibliography 105 [11] A. Pal, “Localization algorithms in wireless sensor networks: current approaches and future challenges.”, Netw. Protoc. Algorithms, vol. 2, no. 1, pp. 45–73, 2010. [12] G. Han, H. Xu, T. Q. Duong, J. Jiang, and T. Hara, “Localization algorithms of wireless sensor networks: a survey”, Telecommunication Systems, vol. 52, pp. 2419–2436, 2013. [13] S. Kumar and R. M. Hegde, “A review of localization and tracking algorithms in wireless sensor networks”, arXiv preprint arXiv:1701.02080, 2017. [14] F. Khan, S. Memon, and S. H. Jokhio, “Support vector machine based energy aware routing in wireless sensor networks”, in 2016 2nd international conference on robotics and artificial intelligence (ICRAI), IEEE, 2016, pp. 1–4. [15] A. Kaur, G. P. Gupta, and S. Mittal, “Comparative study of the different variants of the dv-hop based node localization algorithms for wireless sensor networks”, Wireless Personal Communications, pp. 1–43, 2022. [16] F. Dong, Y. Chen, J. Liu, Q. Ning, and S. Piao, “A calibration-free localization solution for handling signal strength variance”, in Mobile Entity Localization and Tracking in GPS-Less Environnments: Second International Workshop, MELT 2009, Orlando, FL, USA, September 30, 2009. Proceedings, Springer, 2009, pp. 79–90. [17] V. Daiya, J. Ebenezer, S. S. Murty, and B. Raj, “Experimental analysis of rssi for distance and position estimation”, in 2011 International Conference on Recent trends in information technology (ICRTIT), IEEE, 2011, pp. 1093– 1098. [18] P. Barsocchi, F. Furfari, P. Nepa, and F. Potortı, “Rssi localisation with sensors placed on the user”, in 2010 International Conference on Indoor Positioning and Indoor Navigation, IEEE, 2010, pp. 1–6. [19] F. Viani, L. Lizzi, P. Rocca, M. Benedetti, M. Donelli, and A. Massa, “Object tracking through rssi measurements in wireless sensor networks”, University of Trento, Tech. Rep., 2008. [20] H. Obeidat, W. Shuaieb, O. Obeidat, and R. Abd-Alhameed, “A review of indoor localization techniques and wireless technologies”, Wireless Personal Communications, vol. 119, pp. 289–327, 2021. [21] S. Wang, X. Jiang, and H. Wymeersch, “Cooperative localization in wireless sensor networks with aoa measurements”, IEEE Transactions on Wireless Communications, vol. 21, no. 8, pp. 6760–6773, 2022. [22] Y. Zou and H. Liu, “Tdoa localization with unknown signal propagation speed and sensor position errors”, IEEE Communications Letters, vol. 24, no. 5, pp. 1024–1027, 2020. [23] V. Gupta and B. Singh, “Study of range free centroid based localization algorithm and its improvement using particle swarm optimization for wireless sensor networks under log normal shadowing”, International Journal of Information Technology, vol. 12, no. 3, pp. 975–981, 2020. Bibliography 106 [24] H. Ghribi, F. Khelifa, A. Jemai, and M. B. B. Salah, “A review of dv-hop localization algorithm”, in 2021 31st International Telecommunication Networks and Applications Conference (ITNAC), IEEE, 2021, pp. 121–126. [25] P. Tripathy and P. Khilar, “Pso based amorphous algorithm to reduce localization error in wireless sensor network”, Pervasive and Mobile Computing, vol. 100, p. 101890, 2024. [26] X. Shen, B. Xu, and H. Shen, “Indoor localization system based on rssi-apit algorithm”, Sensors, vol. 23, no. 24, p. 9620, 2023. [27] S. Kanchi, “Localization in wireless sensor networks using mds with internode angle estimation”, in Computational Science and Its Applications– ICCSA 2021: 21st International Conference, Cagliari, Italy, September 13–16, 2021, Proceedings, Part I 21, Springer, 2021, pp. 622–630. [28] A. Hadir, K. Zine-Dine, and M. Bakhouya, “Improvements of centroid localization algorithm for wireless sensor networks”, in 2020 5th International Conference on Cloud Computing and Artificial Intelligence: Technologies and Applications (CloudTech), IEEE, 2020, pp. 1–6. [29] H. Will, T. Hillebrandt, Y. Yuan, Z. Yubin, and M. Kyas, “The membership degree min-max localization algorithm”, in 2012 ubiquitous positioning, indoor navigation, and location based service (UPINLBS), IEEE, 2012, pp. 1– 10. [30] X. Zhu, W. Qu, T. Qiu, L. Zhao, M. Atiquzzaman, and D. O. Wu, “Indoor intelligent fingerprint-based localization: principles, approaches and challenges”, IEEE Communications Surveys & Tutorials, vol. 22, no. 4, pp. 2634– 2657, 2020. [31] D. A. Tran and T. Nguyen, “Localization in wireless sensor networks based on support vector machines”, IEEE Transactions on Parallel and Distributed Systems, vol. 19, no. 7, pp. 981–994, 2008. doi: 10.1109/TPDS.2007.70800 . [32] J. Mass-Sanchez, E. Ruiz-Ibarra, J. Cortez-González, A. Espinoza-Ruiz, and L. A. Castro, “Weighted hyperbolic dv-hop positioning node localization algorithm in wsns”, Wireless Personal Communications, vol. 96, pp. 5011– 5033, 2017. [33] W. Zheng, X. Yan, W. Zhao, and C. Qian, “A large-scale multi-hop localization algorithm based on regularized extreme learning for wireless networks”, Sensors, vol. 17, no. 12, p. 2959, 2017. [34] S. Phoemphon, C. So-In, and D. T. Niyato, “A hybrid model using fuzzy logic and an extreme learning machine with vector particle swarm optimization for wireless sensor network localization”, Applied Soft Computing, vol. 65, pp. 101–120, 2018. [35] J. Zheng and A. Dehghani, “Range-free localization in wireless sensor networks with neural network ensembles”, Journal of Sensor and Actuator Networks, vol. 1, no. 3, pp. 254–271, 2012. [36] B. Peng and L. Li, “An improved localization algorithm based on genetic algorithm in wireless sensor networks”, Cognitive Neurodynamics, vol. 9, pp. 249–256, 2015. Bibliography 107 [37] X. Chang and X. Luo, “An improved self-localization algorithm for ad hoc network based on extreme learning machine”, in Proceeding of the 11th World Congress on Intelligent Control and Automation, IEEE, 2014, pp. 564– 569. [38] S. S. Banihashemian and F. Adibnia, “A novel robust soft-computed rangefree localization algorithm against malicious anchor nodes”, Cognitive Computation, vol. 13, no. 4, pp. 992–1007, 2021. [39] L. Wang, M. J. Er, and S. Zhang, “A kernel extreme learning machines algorithm for node localization in wireless sensor networks”, IEEE Communications Letters, vol. 24, no. 7, pp. 1433–1436, 2020. [40] D. Koutsonikolas, S. M. Das, and Y. C. Hu, “Path planning of mobile landmarks for localization in wireless sensor networks”, Computer Communications, vol. 30, no. 13, pp. 2577–2592, 2007. [41] D. Koutsonikolas, S. M. Das, and Y. C. Hu, “Path planning of mobile landmarks for localization in wireless sensor networks”, Computer Communications, vol. 30, no. 13, pp. 2577–2592, 2007. [42] O. Liouane, S. Femmam, T. Bakir, and A. B. Abdelali, “New online dv-hop algorithm via mobile anchor for wireless sensor network localization”, Tsinghua Science and Technology, vol. 28, no. 5, pp. 4849–4854, 2023. [43] G. Han, J. Jiang, C. Zhang, T. Q. Duong, M. Guizani, and G. K. Karagiannidis, “A survey on mobile anchor node assisted localization in wireless sensor networks”, IEEE Communications Surveys & Tutorials, vol. 18, no. 3, pp. 2220–2243, 2016. [44] S. Chai et al., “Localization technology for wireless sensor networks”, Wireless Sensor Networks, pp. 69–141, 2020. [45] S. Messous, H. Liouane, and N. Liouane, “Improvement of dv-hop localization algorithm for randomly deployed wireless sensor networks”, Telecommunication Systems, vol. 73, no. 1, pp. 75–86, 2020. [46] Q. Zhao, Z. Xu, and L. Yang, “An improvement of dv-hop localization algorithm based on cyclotomic method in wireless sensor networks”, Applied Sciences, vol. 13, no. 6, p. 3597, 2023. [47] Y. Cao and J. Xu, “Dv-hop-based localization algorithm using optimum anchor nodes subsets for wireless sensor network”, Ad Hoc Networks, vol. 139, p. 103035, 2023. [48] D. Yang, X. Zhao, and W. Zhang, “A dv-hop localization algorithm using classifying average hop distance in wireless sensor networks”, in Proceedings of the 5th International Conference on Computer Science and Software Engineering, 2022, pp. 203–208. [49] Q. Liang, S. - C. Chu, Q. Yang, A. Liang, and J. - S. Pan, “Multi-group gorilla troops optimizer with multi-strategies for 3d node localization of wireless sensor networks”, Sensors, vol. 22, no. 11, p. 4275, 2022. [50] C. Ge, L. Li, Y. Yu, H. Bian, and Q. Zhang, “Improvement of dv-hop algorithm based on error analysis”, in 2022 IEEE 6th Information Technology and Mechatronics Engineering Conference (ITOEC), IEEE, vol. 6, 2022, pp. 1299– 1303. Bibliography 108 [51] S. Ansari and K. K. Nagwanshi, “An empirical study of soft computing approaches in wireless sensor networks”, Journal of Cases on Information Technology (JCIT), vol. 24, no. 4, pp. 1–10, 2022. [52] L. L. de Oliveira, G. H. Eisenkraemer, E. A. Carara, J. B. S. Martins, and J. Monteiro, “Mobile localization techniques for wireless sensor networks: survey and recommendations”, ACM Transactions on Sensor Networks, vol. 19, pp. 1–39, 2022. [Online]. Available: https://api.semanticscholar. org/CorpusID:252163129. [53] O. Liouane, S. Femmam, T. Bakir, and A. B. Abdelali, “New online dv-hop algorithm via mobile anchor for wireless sensor network localization”, Tsinghua Science and Technology, vol. 28, no. 5, pp. 4849–4854, 2023. [54] O. Liouane, S. Femmam, T. Bakir, and A. B. Abdelali, “Novel dv-hop algorithm-based machines learning technics for node localization in rangefree wireless sensor networks”, Int J Inf & Commun Technol ISSN, vol. 2252, no. 8776, p. 8776, 2023. [55] J. Wang, S. Lu, S. - H. Wang, and Y. - D. Zhang, “A review on extreme learning machine”, Multimedia Tools and Applications, vol. 81, no. 29, pp. 41611– 41660, 2022. [56] C. So-In, S. Permpol, and K. Rujirakul, “Soft computing-based localizations in wireless sensor networks”, Pervasive and Mobile Computing, vol. 29, pp. 17–37, 2016. [57] P. Yadav and S. Sharma, “An efficient optimal localization technique for wsn using hybrid machine learning algorithms”, Wireless Personal Communications, pp. 1–39, 2024. [58] S. - H. Lee, C. - H. Cheng, K. - H. Lu, Y. - L. Shiue, and Y. - F. Huang, “Performance of differential evolution algorithms for indoor area positioning in wireless sensor networks”, Electronics, vol. 13, no. 4, p. 705, 2024. [59] S. Zhang, M. J. Er, B. Zhang, and Y. Naderahmadian, “A novel heuristic algorithm for node localization in anisotropic wireless sensor networks with holes”, Signal Processing, vol. 138, pp. 27–34, 2017. [60] O. Liouane, S. Femmam, T. Bakir, and A. B. Abdelali, “New online dv-hop algorithm via mobile anchor for wireless sensor network localization”, Tsinghua Science and Technology, vol. 28, no. 5, pp. 4849–4854, 2023. [61] R. Sharma, “Optimized clustering using soft computing approaches in wireless sensor networks: research dimensions and contributions”, International Journal of System Assurance Engineering and Management, vol. 13, no. 2, pp. 557–570, 2022. [62] V. Annepu et al., “Review on unmanned aerial vehicle assisted sensor node localization in wireless networks: soft computing approaches”, IEEE Access, vol. 10, pp. 132875–132894, 2022. [63] J. Chadha and A. Jain, “Fuzzy logic-based range-free localization in wsn”, in Machine Learning, Advances in Computing, Renewable Energy and Communication: Proceedings of MARC 2020, Springer, 2022, pp. 89–97. Glossary 115 SVM Support Vector Machine . 28 U UAV Unmanned Aerial Vehicule . 37 W WSN Wireless Sensor Network. 1, 5, 6 X XSG Xilinx System Generator. 4