In allowing us to observe what we cannot touch (and often, cannot see), remote sensing affords the means to look beneath the ground, through the cover of tree canopies, and across great distances through cloud cover. Effective remote sensing must provide its results before the situation changes in the sensed area - hence, the need for real time signal processing arises to quickly extract information from the reams of raw data provided by remote sensors. Furthermore, data bandwidth limitations may dictate that this processing be performed at the sensor, and hence in power-starved, space-limited environments, to extract the limited amount of key information contained within the raw data.
This talk begins by defining some remote sensing concepts that include the three phases of information extraction, combination, and abstraction, then defines the concept of real time signal processing and examines how to achieve real time response upon current computing elements. Topics include algorithm complexity, use of approximate processing, specialized computing architectures, and the use of multiple processors.
Having defined concepts of remote sensing applications and signal processing architectures, the topic of quickly and efficiently mapping a proven application onto a specific algorithm is presented. Methods of algorithm description move from architecture independent to architecture specific, as well as from timing independent to timing dependent. Special emphasis is given to the problem of achieving a rapidly-conceived, non-optimal but real time mapping of algorithm and architecture.
Finally, trends in current remote sensing are described, accompanied by the research questions that they motivate. Research challenges include the autonomous management of a swarm of mobile reconfigurable sensors, accumulating a total picture from the diverse fractionated observations of these sensors, inversion and estimation of complex problems, and self-organizing sensor arrays.
John G. Ackenhusen received the B.S. (Physics), B.S.E (Nuclear Engineering), M. S. (Physics), M. S. E. (Nuclear Engineering), and Ph.D. (Plasma Physics), all from The University of Michigan. He joined Bell Laboratories, Murray Hill, NJ, where he established a group devoted to realizing speech recognition, coding, and synthesis algorithms in real time - this work led to AT&T's first workstation product performing speech processing. He then became Head of the Signal Processing System Engineering Department at Bell Labs, leading the design of the Navy Enhanced Modular Signal Processor for antisubmarine warfare. Greg Wornell has been on the MIT faculty since 1991, where he is Professor in the EECS department. He did his graduate work also at MIT in EECS, and his undergraduate work at the University of British Columbia. His research interests span a variety of aspects of signal processing, information theory, and digital communication, and include algorithms and architectures for wireless and sensor networks, broadband systems, and multimedia environments.
John is the Senior Manager at General Dynamics Michigan Research and Development Center, Ypsilanti, MI, where he leads research in remote sensing methods and processing. He is a Fellow of the IEEE and was President of the IEEE Signal Processing Society in 1990-1991. He is author of the textbook Real Time Signal Processing; Design and Implementation of Signal Processing Systems (Prentice-Hall, 1999).
return to EECS500 Homepage