Mobile Broadband (MBB) networks underpin several essential operations of today’s society by regulating a huge portion of the modern communications system in the world. The recent scientific advances in MBB technologies such as Fifth Generation (5G) and cellular Internet of Things (IoT) will further strengthen the MBB networks’ role, making them the norm in the global mobile telecommunications ecosystem. Given the increasing number of mobile devices and coupled with the high availability of data, it is therefore important to understand the underlying mechanisms that define the behavior of the MBB network performance in the wild. Such constructive feedback is crucial to dictate the network operators’ strategies on on-going or future deployment initiatives. In addition, businesses and application developers can benefit from the research activities by employing performance optimization actions on their services.

In this thesis, we focus on the empirical characterization and modeling of mobile systems. In particular, we are interested in capturing the interplay between numerous network performance metrics, such as bandwidth, latency, and signal strength for MBB networks in the wild. Moreover, we aim to explore the performance and coverage boundaries of the Narrowband IoT (NBIoT) radio technology standard. Toward this goal, we exploit experimental platforms to perform controlled, transparent, and replicable real-world measurements and collect a multitude of attributes and characteristics from operational mobile networks. We further complement our research activities by leveraging crowdsourced measurements, since they constitute a more realistic but rather noisier picture of the mobile network performance in the wild. For the analysis part, we design, implement, and propose supervised learning models using data- driven methods and Artificial Intelligence (AI) paradigms. The research is partitioned across nine papers, including four conference papers, four journal papers, and one demo paper. However, context-wise, it can be grouped under three main categories, i.e., (i) Characterization and data-driven modeling of MBB network performance, (ii) Advanced data analytics for bandwidth forecasting under mobility, and (iii) Dissection of Narrowband Internet of Things (NBIoT) performance in the wild.

First, we focus on the data-driven modeling aspect of Long-Term Evolution (LTE) networks. In particular, we study the impact of LTE parameters on the web application performance as well as speed test measurements. We further propose a proof of concept ML regression based framework to reduce the data volume consumption when running speed test measurements. We complement our research efforts by providing additional insights on the MBB performance leveraging a longitudinal campaign of mobile measurements spanning two countries over a two-year period.

Second, considering the increasing amount of attention on time series performance forecasting applications, such as network traffic management and application provisioning, we study the performance of Long Short Term Memory (LSTM) networks for bandwidth forecasting in MBB settings under diverse mobility. We design an open-source framework that allows experimentation with LSTM networks, including hyperparameter optimization, and we leverage it as a tool to carry out an extensive comparative analysis between 4G and 5G systems.

Last, we perform a large-scale measurement campaign to study the performance of the NBIoT technology in terms of network coverage and deployment. We further analyze the functionality and efficiency of the Random Access (RA) process and provide additional insights, including a Machine Learning (ML) based methodology to predict its outcome. To the best of our knowledge, this is the first publicly available dataset and empirical analysis of NBIoT on operational networks.

Narrowband Internet of Things (NB-IoT) is gaining momentum as a promising technology for massive Machine Type Communication (mMTC). Given that its deployment is rapidly progressing worldwide, measurement campaigns and performance analyses are needed to better understand the system and move toward its enhancement. With this aim, this paper presents a large scale measurement campaign and empirical analysis of NB-IoT on operational networks, and discloses valuable insights in terms of deployment strategies and radio coverage performance. The reported results also serve as examples showing the potential usage of the collected dataset, which we make open- source along with a lightweight data visualization platform.

In this study, we evaluate the performance of different Dynamic Adaptive Streaming over HTTP (DASH) video rate adaptation algorithms over mobile networks. More specifically, we compare and evaluate the performance of Basic, Segment Aware Rate Adaptation (SARA), and Buffer-Based Rate Adaptation (BBA) rate adaptation algorithms over the 4G-Long Term Evolution (LTE) networks of three Mobile Network Operators (MNOs): Telenor, Telia, and ICE. We use the Measuring Mobile Broadband Networks in Europe (MONROE) testbed for conducting measurements, and OpenVQ toolbox for evaluation.