76. ENERGY EFFICIENT DESIGN OF HETEROGENEOUS WIRELESS HEALTHCARE SYSTEMS

Department: Computer Science & Engineering
Research Institute Affiliation: Center for Networked Systems (CNS)
Faculty Advisor(s): Tajana Simunic-Rosing

Primary Student
Name: Priti P Aghera
Email: paghera@ucsd.edu
Phone: 858-534-9892
Grad Year: 2009
Resume: View Resume

Student Collaborators
Diana Fang, dcfang@cs.ucsd.edu

Abstract
Wireless healthcare systems have a set of wireless sensors, one or more wireless aggregators per user and a backend server.The vital signs sensed by the sensors needs to be processed and transferred to the backend server through wireless channels in real time for further analysis by healthcare professionals. One of the key challenges in such a wireless healthcare system is managing energy consumption of the mobile devices used in the system due to its limited and/or irreplaceable battery sources and the need for longer operational time of the system as a whole. A wireless healthcare system consists of a set of sensing tasks that produces the sensed data and processing tasks which processes the sensed data or processed data and communications task which acts as a link between sensing and processing tasks. These set of tasks are associated with each by data producer and data consumer relationship. Assignment of the processing tasks on a node affects node’s battery life due to the associated computation cost and communication cost and task’s latency due to node’s processing and communication speed. In such systems dynamically selecting which processing, sensing and communication tasks should run at what point in time and on what device is difficult, but critically important if a good tradeoff between battery lifetime, latency and high fidelity of sensed and processed data are to be achieved.In this work we explore this trade-off and optimize the battery life of the overall system by optimal assignment of processing task to available resources while meeting latency constraints of the required data at backend. Since task binding and scheduling optimization have been shown to be NP hard, we use the optimal and static ILP formulation of this problem as a baseline comparison with dynamic and heuristic algorithm we developed. Our heuristic adapts to changing conditions in the environment and user’s needs by selecting and binding the tasks to the appropriate nodes in the system at run time with very low overhead. We implemented the proposed solution and show that it improves the battery lifetime of the system significantly while providing a good tradeoff in accuracy and latency.

« Back to Posters or Search Results