University of Minnesota
University Relations
http://www.umn.edu/urelate
612-624-6868
myU OneStop


Go to unit's home.

Home | Seminars and Symposia | Past seminars/symposia: Friday, March 12, 2004

DTC Seminar Series

Information processing in distributed sensor systems: Data acquisition and Inference

by

Prakash Ishwar
Department of Electrical Engineering
University of California at Berkeley

Friday, March 12, 2004
11:00 am

402 Walter Library

Prakash Ishwar

Download presentations (pdf 860 KB) Driven by exciting new applications that include large-scale sensor networks and wireless multimedia transmission, signal processing is undergoing a revolution of sorts. Many of these applications require a shift away from classical centralized architectures and algorithms to more distributed ones. Signal processing cannot be addressed in isolation, but rather as an interdisciplinary system component interacting with areas like communications, coding, information theory, and networking. A range of open problems emerge when classical system tasks are posed in a distributed setting that imposes challenging constraints such as poor precision devices, limited power resources, and poor reliability. This talk will focus on two related functional tasks: (i) information sensing/data-acquisition and (ii) decision making.

Prakash Ishwar

The first part of this talk focuses on sampling bandlimited fields in a distributed, precision-limited, communication-constrained processing environment. The feasibility of having a flexible tradeoff between the sensor density and the A/D quantizer precision while achieving an exponential field reconstruction accuracy in the number of bits per Nyquist-interval is demonstrated exposing an underlying "conservation of bits" principle. This analysis also sheds light on the growth of information with sensor density, available flexibility for sensor deployment, and potential robustness and security features. The final part of this talk briefly explores an intriguing joint classification and compression problem where a complexity-constrained remote sensing unit is incapable of performing the expensive classification task, but can communicate with a more capable central unit through a limited bitrate link. The central unit has access to a database of "templates" which is correlated to the observations of the remote unit. It is shown that in important cases of interest, asymptotically, the classification and rate-distortion performance can be as good as when the remote unit has direct access to the database. Applications include low-complexity video coding and security.

 

Prakash Ishwar received the B. Tech. degree in Electrical Engineering from the Indian Institute of Technology, Bombay, in 1996, and the M.S. and Ph.D. degrees in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign in 1998 and 2002 respectively. Since August 2002, he has been a post-doctoral researcher in the EECS department in the University of California at Berkeley. His professional interests include distributed/collaborative signal processing, multiterminal information theory, statistical/information-theoretic image and video modeling and processing, and detection and estimation theory. He has served on the Technical Program Committees of ICIP 2003 and IPSN 2004 and was a co-organizer of the 2003 Berkeley Mini-workshop on the Fundamentals of Sensor Webs. He was awarded the 2000 Frederic T. and Edith F. Mavis College of Engineering Fellowship of the University of Illinois.