Abstract
5G (Fifth generation) networks are based on the next-generation wireless networktechnology which is expected to change the way people live and work. Localization is
considered a key component in the 5G network, especially for location-based services. It infers predicting an object’s location in a given environment. Providing information about the physical location of a user is a highly desirable feature of future wireless networks. 5G's millimeter-wave transmissions bring a new paradigm to wireless communications, as it allows high-speed wireless communication on frequencies between 30-300 GHz. This generation of wireless networks utilizes high-frequency radio waves which have significantly improved localization. Moreover, it has proved that we can achieve the best localization accuracy with minimum cost and its accuracy is reliable. We use highfrequency radio waves like mmwave in outdoor localization. Due to good signal bandwidth of mmWave, we achieve good accuracy in high-frequency localization. In this project, we propose a deep learning based 5G mmWave localization using Transformer Neural Network Architecture. The main aim of this project is to minimize the location error of the user leveraging transformer, neural network-based architecture, which utilizes the mechanism of attention. The model is trained and validated on DeepMIMO Dataset (A Generic Deep Learning Dataset for Millimeter Wave and Massive MIMO Applications). The project has been implemented using hardware accelerators like GPU in a python-based framework called TensorFlow. The dataset has been generated using MATLAB and Google Colab Pro has been used to train the models for the generated dataset [14].