Energy-driven Optimization of Hardware and Software for Distributed Embedded Systems

Distributed embedded computing systems are special-purpose computer systems designed for particular applications and set up in a networked or distributed manner. A practical application domain for such a distributed system setup is the domain of wireless sensor network (WSN) applications. In this thesis, studies of architectures, applications, and methodologies for distributed embedded systems will be covered by addressing a number of important energy and performance optimization problems for translating highlevel representations of distributed embedded applications into system implementations. This thesis is also concerned about systematic design methodologies and optimization problems for both software and hardware implementations. With advances in integrated circuit technology, distributed embedded platforms such as wireless sensor nodes can be equipped with increasing amounts of computational resources, such as digital signal processing (DSP) subsystems that can handle intensive Title of dissertation: ENERGY-DRIVEN OPTIMIZATION OF HARDWARE AND SOFTWARE FOR DISTRIBUTED EMBEDDED SYSTEMS Chung-Ching Shen, Doctor of Philosophy, 2008 Dissertation directed by: Professor Shuvra S. Bhattacharyya Department of Electrical and Computer Engineering, and Institute for Advanced Computer Studies computational tasks. By incorporating such dedicated DSP processing, a distributed embedded platform can enhance its functional capabilities for processing data before transmitting the data to other parts of the network or to an associated base station (central node). Therefore, this thesis presents a design methodology for distributing DSP applications across wireless sensor networks and optimizing associate trade-offs between computation and communication. A low-power, application-specific sensor node platform for distributed WSN systems is designed and demonstrated in this thesis. This platform provides mixed-signal integration of streamlined digital circuits for protocol control and data processing, along with required analog subsystems, such as transceiver circuitry. Building on this optimized platform, this thesis demonstrates a complete system design of an application-specific WSN system with compact size and low power features. This system design is the result of an integrated effort across design space exploration, algorithm development, crosslayer protocol design, and most importantly, the completion of various hardware prototype implementations for validating and demonstrating proposed design techniques. This thesis also presents a system-level synthesis methodology for finding the most suitable resource configurations for distributed, embedded systems. System-level synthesis is attractive because the carefully designed system-level models can be analyzed and evaluated rapidly, and the complex, inter-related design decisions can be explored and evaluated at a high level before mapping into low level implementations. We demonstrate the accuracy and efficiency of our system-level synthesis approach, and its ability to capture an important range of high level interactions that are relevant to the design of distributed, embedded systems. ENERGY-DRIVEN OPTIMIZATION OF HARDWARE AND SOFTWARE FOR DISTRIBUTED EMBEDDED SYSTEMS