IoT solutions for e-Health applications for care’s continuity at home

This research investigates the use of Channel State Information (CSI) extracted from Wi-Fi signals as a fine-grained and physically interpretable observable for contactless monitoring of human activity, breathing, and environmental characteristics. The study builds on the premise that the complex amplitude and phase of the wireless channel, traditionally employed for communication diagnostics, also contain rich information about subtle propagation variations induced by materials, objects and human motion. The thesis is organised into five main chapters. The first chapter introduces the theoretical background and motivation, highlighting the growing interest in device-free sensing and the potential of CSI as a bridge between communication and perception. The second chapter reviews the state of the art, tracing the evolution of CSI-based sensing techniques from amplitude-only analysis to phase-calibrated and deep-learning-driven approaches. The third chapter details the data acquisition methodology, the MATLAB simulation tools, and the signal processing pipeline. The fourth chapter presents the experimental and simulation results across three key domains: (i) material and object identification, (ii) recognition of human movements and activities, and (iii) monitoring of the breath rate. Finally, the fifth chapter provides the general conclusions of the research and outlines the directions for future work. Under controlled line-of-sight (LoS) conditions and with single subjects, the results demonstrated high stability and temporal coherence. Traditional machine learning models such as Support Vector Machines (SVM) and Random Forests achieved accuracies above 95% in material and object classification. Deep learning architectures, including Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, surpassed 97% accuracy in movement recognition and breathing detection. These results confirmed that CSI encapsulates a dense, multidimensional representation of the radio channel, capable of capturing both macroscopic motion and sub-centimetric physiological oscillations. However, experiments also revealed a degradation of performance in non-line-of-sight (NLoS) and multi-subject scenarios, where multipath interference and temporal non-stationarity reduce amplitude and phase coherence. This limitation motivated the exploration of Reconfigurable Intelligent Surfaces (RIS) as a means to actively shape and strengthen wireless propagation. A MATLAB-based RIS simulator was implemented to model programmable reflective arrays operating at 28 GHz, where the phase of each element is adjusted to steer or reinforce the signal toward the receiver. Simulations varying both the number and position of RIS elements showed signal-to-noise ratio (SNR) gains of up to 20 dB in obstructed configurations, confirming the RIS’s ability to restore phase stability and enhance CSI quality in challenging propagation conditions. The final chapter discusses the broader implications of this research and identifies future directions. The integration of RIS technology represents an initial step toward adaptive wireless environments capable of dynamically controlling propagation to improve sensing accuracy. Future work will include experimental validation of RIS prototypes in real environments, the adoption of ESP32-based Wi-Fi hardware for full OFDM subcarrier analysis, and the development of hybrid analytical–data-driven models combining physical channel formulations with machine and deep learning inference. In the long term, the aim is to move toward intelligent, self-optimising wireless systems where communication and perception are seamlessly unified.