An Investigation of data-driven Methods Applied to Telecommunications

The main objective of this thesis is to investigate the impact of Data-Driven techniques in ICT Engineering, with a specific focus on telecommunication systems. Firstly, a brief background on Data-Driven methods is provided, containing a recap on Machine Learning and Data reduction methods. Then, a literature review illustrates the impact of Data-driven methodologies on applications in the field of communication systems. The contributions of this PhD work are related to 4 different engineering applications to telecommunication systems, and are described as follows: (1) the use of data-driven methodologies to improve Multiple-Input Multiple-Output (MIMO) performance for crosstalk cancelation in optical systems supporting multiple spatial modes is addressed. In this respect, we proposed a reduction algorithm based on the Principal Component Analysis (PCA) and cross-correlation analysis to improve traditional equalizers’ performance. (2) A novel regression trees-based methodology able to learn a Markov model of a fading channel via historical data of the signal-to-interference-plus-noise-ratio (SINR) is proposed. Such methodology is used to derive a Markov jump model of a wireless control network and thus to design a stochastic optimal controller that considers the inter-dependence between the plant and the wireless channel dynamics. Our methodology is validated using a WirelessHART point-to-point communication based on the IEEE-802.15.4 standard. (3) A novel complexity reduction methodology is proposed for a data-driven control algorithm based on regression trees. In particular, the refinement procedure aims to reduce the dimension of the dynamical model without compromising the model accuracy and mitigating the overfitting problem. (4) the application of edge computing for real-time analysis to support autonomous operations of unmanned aerial vehicles (UAV) is addressed. Indeed, AUV autonomous operations necessitate the real-time analysis of information-rich signals, such as camera and LiDAR feeds, where the analysis algorithms often take the form of extremely complex deep neural networks (DNN). The continuous execution of such models onboard the UAV imposes a considerable resource consumption (e.g., energy), while offloading the execution of the models to edge servers requires the transmission of the input signals over capacity-constrained, time-varying, wireless channels. We propose an innovative approach to control the computing pipeline of signal processing.