Human body-enabled methods for the implicit synchronization of distributed wearable systems

During the last three decades, wearable computing has experienced an impressive evolution. While single all-round devices, worn at familiar and convenient locations such as the wrist, have been standard for many years, wearables are recently evolving rather into a conglomerate of simultaneously deployed, specialized sensing units that can be attached to various sites with complementary perspectives. Applying cutting-edge sensor fusion and machine learning techniques on the collected multi-device data allows inferring the user's bigger picture with an accuracy superior to that from a single site. However, the applied techniques suffer significantly from the inaccuracies of the devices' internal clocks, the manual temporal alignment of their sensor channels, and hence the recordings' unmatched time bases. The available radio-based online or event-based offline synchronization methods either considerably affect the devices' battery life or often do not achieve the required accuracy. Moreover, both methods demand for the user's explicit interaction to either network the distributed devices or perform specific synchronization actions. This dissertation hypothesizes that attaching devices to the human body can enable technologies and methods that would not be possible without its presence. Two implicit methods are presented to address the lack of accurate and efficient synchronization principles for distributed wearable systems. Therefore, two perspectives are investigated in which the human body is considered first as a source of natural signals and second as a transmission medium to provide artificial signals throughout the body surface. The transfer of the medical sensing modalities electrocardiography and photoplethysmography to a wearable form factor has enabled constant access to the user's vital signs. The first method PulSync leverages the irregular rhythm of the heartbeat, ubiquitously and simultaneously detectable throughout the entire body surface. Modulated by various physiological processes, the inter-beat intervals form patterns in the derived heart rate variability interval function that are unique like a fingerprint and, therefore, can serve as significant landmarks for the offline alignment of recordings in a data-driven post-processing step. The novel communication principle of intra-body communication is somewhat located between traditional wired and wireless techniques while showing advantages over both. The second method IBSync is based on artificial landmark signals that are either consciously or implicitly and incidentally induced into the user's skin by touching or passing areas or objects equipped with transmitter beacons. Obtained from the augmented human body, the detected landmarks are enriched with data, allowing to uniquely align the recordings offline or even allocate landmark positions in the absolute time. Both methods achieved an alignment accuracy in the order of a single sample and show a comparable performance: PulSync with -0.714 ± 3.440 samples and IBSync with 0.800 ± 1.792 samples. Therefore, with the achieved temporal accuracy of 2.86 ms at 250 Hz and 6.25 ms at 128 Hz respectively, the two methods PulSync and IBSync are superior to most available offline synchronization methods and can even keep up with common online methods.