| For decades, we have observed Moore's law in action, shrinking digital electronic devices and endowing them with more and more computing power. Simultaneously, interest increases in the design of protocols, hardware and software components, and applications for networked embedded systems. The use of wireless communication links and the addition of sensing and actuation devices has encouraged their use in many domains in science and industry. A broad scope of applications and devices that provide health and medical services, often termed telehealth, wireless health, or medical monitoring, have benefitted from these developments and are an very active area of research.;Architects of these systems are forced to consider the data to be collected and the end-user experience. Where the embedded system design is concerned, optimization for resource conservation---for example, energy dissipation or network bandwidth---can be performed independently of the characteristics of the data being collected. Building and optimizing models of the signal measurements is paramount when scientific investigation is relevant. Models can be optimized for their accuracy in representing patterns present in the data, and their complexity can be reduced to make them efficient when they are used in production systems. In the quest for simpler models, we hope to help find simpler scientific explanations for the phenomena under study.;This dissertation considers challenges existing in both the signal processing and networking domains. In both cases there are obstacles to be met with methods to efficiently aggregate over a collection of objects of interest; in the networked embedded system design case, we consider efficient aggregation of data measurements made at individual nodes, while in the data modeling case, we consider efficient and accurate methods to aggregate, or group, variables in models. We develop and test protocols to allow arrays of wireless-networked, medical-monitoring nodes to scale gracefully when they communicate, and we also present model selection techniques that facilitate more accurate and precise models, specifically for medical monitoring systems and their associated physiological signals. |
