System Design and Signal Processing to Facilitate Health Monitoring via Wearable IoT Devices

By 2050, the European Union expects a significant reduction in people under 60 and a surge in those over 65, leading to an inevitable strain on healthcare systems due to more age-related diseases and care requirements. Long-term health monitoring allows for an early diagnosis of diseases and more effective treatment, but its potential has been limited since it currently revolves around costly, inconvenient visits to dedicated facilities in the present-day, physician-centered healthcare system. Wireless sensor networks of low-power wearable Internet of Things (IoT) devices may now enable accessible, cost-efficient, and unobtrusive health monitoring, thereby unburdening some of the healthcare systems' scarce resources. However, to make this promising vision a reality, several open problems must be addressed. For one, the required wearable IoT devices for health monitoring are envisioned to rely on very little power and moreover, a seamless integration into daily life of the intended users is necessary. Wireless power transfer (WPT) facilitates smaller, less obtrusive devices. The main contributions for facilitating reliable WPT to low-power wearable IoT devices in typical indoor environments are summarised in the following. Optimal Antenna Deployment and Power Allocation for Reliable WPT: Multiple-input single-output (MISO) WPT systems under line-of-sight (LoS) conditions are considered, whereby energy receivers (RXs) may lie in the radiating near-field of the transmitter (TX). First, a single-user two-antenna WPT system is considered, whereby perfect channel state information (CSI) at the TX is assumed. The amplitude variations in the wireless channel are exploited to optimise the transmit antenna placement, which is derived in closed form, to achieve a reliable provision of power by maximising the lowest received power in the considered environment. In general, a co-located transmit antenna architecture is shown to be optimal in the far-field region and, for most environment geometry's, a distributed antenna system (DAS) is the optimal deployment solution for the radiating near-field region. Next, a multi-user multiple-input single-output (MISO) WPT system intended for supplying an indoor environment with a guaranteed power level without CSI is considered. For the TX, a continuous aperture antenna is employed. The continuous transmit power distribution of the TX is optimised in order to maximise the lowest received power in the room. Analytically, a continuous aperture antenna is shown to be unnecessary in the context of the considered problem, and in fact, a finite number of radiating elements is proven to be optimal for a one-dimensional deployment region. The optimal transmit antenna deployment and power allocation is determined numerically by discretising the transmit power distribution's domain. Hereby, deploying a finite number of transmit antennas, which radiate a fixed amount of power, is found to be optimal, in general. Beyond WPT, wireless systems for health monitoring via wearable IoT devices must support other functionalities as well, such as sensing the environment for localising RXs, for example. Efficient resource utilisation, such as spectrum use, is crucial as the number of (wearable) IoT devices is expected to grow significantly to enable novel IoT applications. The main contribution for facilitating an efficient co-design of WPT and sensing functionalities is summarised in the following. ISAPT System Design: Integrated sensing and power transfer (ISAPT) system design is investigated for maximising the weighted sum of average harvested powers at RXs while ensuring a guaranteed range estimation accuracy in a certain direction. A low-complexity algorithm, which converges to a stationary point, based on a one-dimensional grid search over the pulse duration in conjunction with semidefinite relaxation (SDR) and successive convex approximation (SCA) is proposed. Thereby, the transmit beamforming vector and the signal amplitude are obtained for a given pulse duration. Taking the energy harvesting (EH) non-linearity at the RXs into account is found to be essential for achieving efficient ISAPT. Furthermore, the trade-off between the functionalities of the ISAPT system is investigated, whereby less power can be harvested for a higher sensing accuracy requirement. Data acquired by health monitoring devices provides insight into the physical and mental state of a human, beyond basic vital parameters. Based on the example of human movement, this thesis proposes a novel method to extract neural activity from muscle activity via electromyography (EMG) signals. This is summarised as follows. Unsupervised Decomposition of Multi-Channel EMG Signals: A fully unsupervised neural network approach is proposed for decomposing multi-channel EMG signals. The neural network, implemented as an autoencoder, is designed for learning the connection between muscle activity and the underlying neural activity from the central nervous system (CNS). In contrast to state-of-the-art methods that linearly separate the muscle activity into motor units (MU), the proposed approach is inherently non-linear due to activation functions employed in the neural network. The proposed decomposition method is validated on synthetically generated EMG data as well as on experimental intramuscular EMG data recorded from the tibialis anterior muscle. On experimental intramuscular EMG data, the proposed method is shown to outperform a state-of-the-art decomposition method in terms of the number of reliably identified MUs.