چكيده به لاتين
Localization has been studied for decades and many solutions have been proposed for it. However, there is a growing demand for location-based services due to its wide applications. Among the areas of interest for researchers and industry at this time are navigation, tracking, surveillance or internal security, law enforcement, quick response in emergency situations, preventing collisions between cars and pedestrians and many other cases. Location information is now an important feature in most Internet of Things (IoT) applications. The idea of localization with the availability of Big Data and using AI-based methods is considered as an important component of future wireless networks, so that the location of anything plays an important role in improving most IoT-based services. Positioning attempts to locate mobile or stationary devices (including smartphones, drones, wearable smart devices such as watches or any other smart element mounted on objects or the environment, and vehicles) using special anchor nodes and mobile devices computing capabilities. Recently, the implementation of new technologies in the field of wireless networks such as 5G and its high popularity, have made it more important to focus on localization methods with the infrastructure of telecommunication networks. This thesis attempts to address this topic by focusing on methods based on the use of the received signal strength on the user’s device in telecommunication networks in outdoor areas, in addition to considering the use of modern tools such as methods based on fingerprints and artificial intelligence. In order to increase accuracy and expand the range of evaluable parameters in the large dataset, we have studied and reviewed past researches in this field. We have implemented a machine learning method on the data we have collected in Tehran. For this purpose, the data was gathered and labeled by drive test method in an Android application designed and developed for this purpose. In the deep learning method used, this dataset has been used to predict the location of the current base station connected to the user’s device. With the results obtained after applying data preprocessing methods and setting hyperparameters, the average distance error predicted in this work has finally been reached to 20 meters. Furthermore, the most accurate model had 99.5% accuracy.