Machine learning for localization using 5G signals

Abstract

5G is the fifth generation of wireless communication technology that offers faster speeds, lower latency and more reliable connections than previous generations. The introduction of 5G mmWave transmission marks a new era in wireless communication. The issue of localization is now compounded by the emergence of intricate and non-linear phenomena like reflections and scattering, which were previously only attributed to signal attenuation, due to the presence of physical obstacles. The encountered obstacles shape the propagation environment, resulting in a multipath effect and indicating the presence of substantial concealed spatial information within the received signal. Meanwhile, accurate and efficient modeling of indoor radio propagation is crucial for designing and operating wireless communication systems. Recently, researchers have explored combining radio propagation solvers with machine learning (ML) to enhance the accuracy and efficiency of these tools. These attempts hold great promise for advancing the state-of-the-art in indoor radio propagation modeling. In order to obtain the hidden information contained in the received information for positioning, this dissertation uses the fingerprint dataset constructed by the received CSI (Channel State Information) signal, and uses different types of machine learning methods for positioning, including traditional CNN (Convolutional Neural Network), LSTM (Long Short-Term Memory) and TCN (Temporal Convolutional Network) model that can predict sequences. Experiments suggest that the deep learning method has a good performance in complex problems such as 5G signal positioning, and the time series-based model has a higher accuracy rate, laying the foundation for the further application of deep learning in this field.