Sensors provide the fundamental link between the computation and physical domains. Although sensors and measuring devices have existed since the early days of science, miniaturization, integration, wireless technologies, and advances in computing are paving the way for developments of very large scale networks of sensing devices, embedded unobtrusively in a wide array of environments. Under this new emerging paradigm, reliability, accuracy, and long term autonomous operation become vital factors in the design and use of practical systems. The intellectual merit of the proposed activity: The goal of a sensing system is often to detect, observe, and learn things about the environment, involving tasks such as signal detection, estimation, classi.cation, or other data analysis operations. These tasks are especially challenging in networked sensor systems for the following reasons: First, it is di.cult to collect data at arbitrary points in time and space as needed. The restricted nature of the measurement process can place severe limitations on the achievable accuracy of inference schemes. Second, calibrating sensor network systems is extremely di.cult due to their decentralized nature. Third, sensor network data analysis tasks can be ill-posed because of the limitations on sampling and calibration. Fourth, key resources such as power, bandwidth, and computing devices are often very limited. This places severe restrictions on the amount of communication and computation that can take place in a sensor network. Detection, estimation, classi.cation, calibration, and communication, each, in their own right, pose very challenging new problems in the context of large-scale sensor networks. Moreover, as a consequence of the four issues raised, distributed sensing, processing, and communication are fundamentally interconnected. Our approach to this challenging new domain rests on investigating the following key principles: distributed algorithms, complexity management, sensory feedback, and the seamless integration of sensing, processing, and communications. Distributed Algorithms for in-network data processing and communications eliminate the need to transmit raw sensor data to a central point. They can provide signi.cant reductions in the amount of communication and energy required to obtain an accurate estimate. Furthermore, distributed algorithms are much more robust and reliable in the presence of communication errors, network outages, and component degradation. Complexity Regularization provides theory and practical methodologies for coping with trade-o.s between model complexity/.exibility and over.tting to limited data. Sensor Feedback allows for dynamic resource allocation, enabling potentially dramatic gains in the use of precious resources such as power, bandwidth, and data processors. Integration of sensing, processing and communication operations is critical for reliable, accurate, and long-term environmental monitoring using networks of sensors. This project will develop a new theory and a suite of algorithms for practical sensor network systems. The algorithms will be analyzed theoretically and tested experimentally. The testbed will consist of optic, acoustic, thermal and other sources and sensors, networked using IEEE 802.15.4 compliant nodes. The combination of these sensors will enable in depth exploration of complex issues including the characterization of spatio-temporal environmental variation, multimodal sensing, and the integration of sensing, data processing, and communication. Broader Impact: This project will provide a unique platform for training graduate and undergraduate students by exposing them to a broad variety of challenges in designing networked sensor systems from both experimental and theoretical perspectives. The topics of the proposed research are an integral part of a new graduate level course o.ering by the PI at the University of Wisconsin Madison Electrical and Computer Engineering Department dealing with networked embedded systems (ECE 902). Furthermore, involvement of minorities and women in the project will be encouraged via various existing programs at the University of Wisconsin. Currently three female graduate students are being supervised by one of the PI's. This project will also facilitate collaborations with local industry who have expressed strong interest and committed to furnishing the hardware and software components necessary in build the experimental testbed.
