Exploiting the Dynamically Architectural Configurability for Compressed Sensing
Sensors or sensing systems are increasingly critical in a variety of applications including national security, surveillance monitoring and health care. Those systems should function with minimal hardware recourses, minimal communications and minimal computation overhead, and these efficiencies can dramatically improve the performance, reliability and usability, which can broaden the overall application scope of sensor systems. This project is to pursue preliminary results of dynamic configurability of architectural and circuit models in sensing systems, and the proposed research will have significant impacts on a range of sensing applications under the resource-constrained environment. For example, in large-scale sensor networks or implantable sensors, energy is tightly constrained. The ultimate goal of the research is to exploit the configurability and dynamics of sensing systems to improve the overall system efficiency. This project serves as an expedition to investigate the dynamically architectural sensing techniques and may open a new research direction of theory and practice in the signal acquisition. Upon the success of this project, a better performance-energy tradeoff in the sensing system will be obtained, which can further strengthen its advantage compared to other sampling techniques, and extend its application regime.
Specifically, this project investigates the dynamic configurability of parameterized Compressed Sensing architecture. With the physical and architectural models, the Compressed Sensing architecture is flexible and provides a larger design/configuration space, and can adapt towards different signal structures and use conditions. The research work is expected to explore a deeper bound of the performance-energy by exploiting the architectural configurability with physical models. To this aim, a set of research tasks will be performed in this project, and the technical thrusts can be summarized from three aspects. First, the project will explore the configurability at both architectural- and circuit-levels in Compressed Sensing, incorporating signal structure variations. Multiple factors in the Compressed Sensing will be investigated. Second, by integrating physical models into the Compressed Sensing architecture, a larger design space will be discovered and defined. The benefit of the performance-energy trade-off will be demonstrated in the new space. Third, a set of novel algorithms will be developed for efficient configuration search in the design space. Several deterministic and heuristic strategies will be investigated in the project.