6533b82efe1ef96bd12923c4

RESEARCH PRODUCT

Technics Amélioration for geo-localization in wsn via smart computing and real time application

subject

description

New technologies exploiting digital information acquisition by radio frequency techniques are now commonly used in various practical fields. They are most often used to measure a variety of physical variables such as temperature, humidity, speed, etc. and are gathered under the name of Wireless Sensor Networks “WSN”. For this variant of applications, the accurate location of connected sensor nodes remains an important issue for researchers and industrial applications. Indeed, existing localization algorithms can be classified into two categories known as « range-based » and « range-free ». Range-based localization systems are characterized by major drawbacks. The first one is the cost of the additional hardware required to measure the distances between the sensor nodes. The other disadvantage concerns the accuracy of the measurements, which can vary according to several parameters related to the nature of the network and the environment: humidity, electromagnetic noise, obstacles, etc. In practice, range-free WSN exploits the notions of connectivity and hop count between inter-nodes to effectively avoid these two drawbacks. Indeed, the fixed nodes of the sensor network whose positions are known are called « anchors ». The other nodes subject to localization process with unknown positions are called « normal nodes ». To estimate their positions, these normal nodes first collect connectivity information from the network as well as the positions of the anchors, and then compute their own positions without the addition of extra hardware for distance measurement and evaluation. Range-free WSN can therefore be adapted to any type of wireless transmission.The objective of this thesis is to perform a study on the localization problem in wireless sensor networks as well as the different tools used for the « range-free » family. The points of study are located at the level of the localization algorithm of type « Dv-hops » and the proposal of a new technique of improvement of the localization precision via the machine learning tools known as "Smart Computing" based "Extreme Learning Machine (ELM)" as well as the implementation of the model of the localization on an FPGA hardware reconfigurable architecture. The thesis is organized as follows: Firstly, we present the different advances in wireless sensor networks and their recent applications in the emerging domains of "IoT" and "Industry 4.0". Secondly, we describe the methodology adopted for localization in range-free wireless sensor networks. Indeed, the multi-layer Extreme Learning Machine is proposed to improve the localization accuracy in wireless sensor networks. A comparative study between the localization results is conducted involving the DV-Hop algorithm, the single hidden layer ELM and the Deep-ELM characterized by two hidden layers. Finally, we describe the different phases of implementation of our localization approach in wireless sensor networks via machine learning and especially the two hidden layer ELM on an FPGA hardware architecture. The software and hardware implementation tools used are the "Matlab-XSG" from Xilinx for the simulations and the generation of the VHDL codes, and the Vivado-HLS tool for the synthesis and the implementation on FPGA. Finally, conclusions and perspectives of our work are presented.