Data-Semantics Driven Power Optimization of Wireless Medical Monitoring Devices

We study techniques to improve deployment of remote wireless health monitoring Body Area Networks (BANs) that offer the promise of improved quality and availability of healthcare. Ease-of-use, safety and security have been sighted as key problems to be resolved before the scale and duration of their deployment can be extended. We focus on 4 performance factors that influence these issues: (i) Semantic Fidelity - The diagnostic quality of data measured by BANs is critical to system safety; (ii) Low-Power Operation - Reducing power consumed by BANs will result in light-weight systems with small form-factors and extended lifetimes, owing to reduced battery discharge rates and the use of smaller batteries; (iii) Fault-Tolerant Operation - Faulty sensor measurements can severely impact diagnostic quality of the data produced by the system, thereby impacting system safety; (iv) Data Sharing Security - Sharing large BAN collected medical datasets exacerbates problems related to secure control over medical data as it is shared between data owners and consumers. Here, we study low-power sampling techniques for multi-sensor remote health monitoring BANs, where medical diagnostic metric accuracy is maximized while the number of samples taken simultaneously is constrained. Power usage is reduced both due to the decrease in number of sensors that are active at a time, and due to the reduced bandwidth required to transmit the measured signals. The techniques are extended to admit fault tolerance and fault detection capabilities while maintaining low-power operation and high semantic-fidelity requirements. We study low-power sampling for the continuous monitoring and anomaly detection use-cases. The use of a novel energy harvester technology in scavenging energy from foot-strikes and powering BANs in a self-sustaining manner is also studied. Several optimizations are considered to maximize the energy output of a proposed foot-strike energy harvester platform based on this novel harvester, while adhering to user-comfort requirements.Finally, we study a robust watermarking technique to address secure data control requirements related to sharing BAN generated datasets. Three use-cases are addressed - proof-of-ownership, data tracking and content authentication. It is shown that robust and imperceptible watermarking of biomedical signals is achievable without compromising semantic fidelity of the datasets