Continuous Physiological Sensing Outside Controlled Environments
Manuel Meier
PhD Thesis, ETH Zürich, 2025
Physiological monitoring plays a crucial role in precision medicine to enable the continuous assessment of an individual's health status through real-time physiological and behavioral data. Among the most widely monitored vital signs is heart rate, which is relevant in a broad range of contexts such as detecting cardiovascular conditions, tracking exercise intensity, or evaluating sleep quality. Other vital parameters, such as respiratory rate, also provide important insights and can serve as early indicators of adverse health events. However, studies inside controlled laboratory or clinical settings fail to capture the complexities of human physiology that occur in free-living conditions. Wearable sensors offer a promising solution to real-world physiological data collection but face challenges such as fragmented and noisy data recording, the need for user interaction, sensor synchronization, and energy constraints that hinder long-term operation. Combined, these challenges limit the effectiveness of wearable technology for truly passive, accurate, and scalable health monitoring. In this dissertation, we introduce an embedded wearable sensing platform and associated computational techniques for robustly monitoring physiological metrics. Our first contribution is the design of a multi-device, low-power sensing system that integrates multiple physiological and environmental sensors, which ensures longduration, passive data collection without requiring user interventions. Our platform achieves sample-perfect synchronization across sensors and millisecond-level alignment between devices, overcoming one of the key barriers in multi-modal physiological monitoring. To further enhance the reliability of physiological data collection, we propose a multi-modal offline synchronization method that enables precise alignment of signals across multiple body-worn sensors without requiring wireless communication or explicit user actions to support real-world scenarios that cannot guarantee continuous connectivity. We demonstrate the capabilities of the platform by introducing WildPPG, a large-scale real-world photoplethysmography (PPG) dataset collected from 16 participants over 13.5 hours in highly variable environmental and activity conditions. Unlike datasets captured inside controlled settings, WildPPG contains physiological responses to real-world activities such as physical exertion, temperature fluctuations, and varying ambient light levels. This makes our dataset a valuable resource for developing more robust wearable sensing algorithms. iii Based on WildPPG, we propose a series of signal-processing techniques for robustly tracking physiological key metrics. We present a lightweight sensor fusion method that enhances heart rate estimation by combining multiple PPG signals from different body locations, mitigating the effects of motion artifacts and environmental disturbances. We also introduce a learning-based approach for generating PPG signals using multiple optical wavelengths, improving cardiac monitoring in multi-wavelength PPG sensing devices such as pulse oximetry applications. Finally, we propose a novel method for continuous respiratory rate monitoring using ultra-wideband (UWB) radar integrated into wearable devices, enabling accurate respiration tracking with reduced susceptibility to motion artifacts. By integrating hardware innovations with computational methods, this dissertation establishes a foundation for continuous, multimodal, and unobtrusive physiological sensing in real-world environments. The wearable system and data processing techniques we propose can advance precision medicine via the passive health monitoring needed to collect representative data for the purpose of early disease detection, long-term physiological research, and, more generally, contextaware health monitoring solutions.
@phdthesis{meier2025,
author = {Manuel Meier},
title = {Continuous Physiological Sensing Outside Controlled Environments},
school = {PhD Thesis, ETH Zuerich},
year = {2025}
}
