Location Awareness in 5G and B5G Ecosystems: Characterization, Design, and Experimentation

Location awareness is a key enabler for a variety of verticals and use cases (UCs) in 5th generation (5G) and beyond 5G (B5G) networks, including those related to autonomy, logistic, smart environments, and Industry 4.0. However, fulfilling the key performance indicator (KPI) requirements for such UCs is challenging. This calls for new localiztion algorithms able to learn from the environment and to fully leverage the positional information provided by the network measurements. Moreover, the integration of next generation cellular networks with sensor radar networks (SRNs), will be fundamental to further enhance these new verticals, as well as to improve the communication performance and the network resource management. This calls for an accurate modeling of the wireless impairments and the design of algorithms able to provide physical analytics (e.g., number of person in a monitored area) in addition to location information. The main objectives of this thesis are: 1. design of machine learning based algorithms for localization in 5G and B5G networks; and 2. characterization of wireless impairments in SRNs, as well as the design of algorithms for extracting physical analytics via SRNs. In particular, this thesis presents the design of soft information (SI)-based localiza- tion algorithms exploiting both radio access technology (RAT)-dependent (obtained from the 5G network) and RAT-independent (obtained via non-3rd Generation Partnership Project (3GPP) technologies) measurements. Performance using both SI and classical approaches are quantified in 3GPP standardized scenarios via rigorous simulations in full conformity with 3GPP technical specifications and reports. Results show that the pro- posed SI approach significantly outperforms the approaches reported in 3GPP technical reports. In addition, a statistical characterization of the clutter for SRNs employing ultra-wideband (UWB) signals is provided based on experimental measurements carried out in an indoor environment. Lastly, a crowd-centric counting algorithm based on machine learning techniques is proposed and compared with state-of-the-art approaches based on experimental measurements.