An Information Theoretic Viewpoint to Performance Bounds of Feedback Systems:  Optimality Results and Open Problems

 

Professor Nuno C. Martins

University of Maryland – College Park

Department of Electrical and Computer Engineering

 

In this seminar, I will discuss recent results on universal  bounds of performance for feedback systems. By using  notions, from Information Theory, such as mutual information and (differential) entropy, I characterize conservation laws that hold under causality, which is a basic attribute of physical systems.  In particular, I show that the differential entropy, induced by external excitation, cannot be reduced by feedback. This principle is related to the Bode integral formula, originally derived for linear and time-invariant feedback systems. The aforementioned analysis extends BodeÕs ideas to arbitrary feedback, while providing a suitable framework for considering the case where a finite-horizon preview, of the external excitation, is available. In addition, I will describe an information flow inequality that can be used for establishing a universal bound of performance, in the frequency domain. These results are expressed as bounds, which, holding in great generality, are useful in establishing the optimality of certain feedback schemes.

 

Bio: Nuno C Martins received the Licenciatura and the MS. degrees in electrical engineering from Instituto Superior Tecnico, Lisbon, Portugal in 1997 and the Ph.D. degree in electrical engineering and computer science from Massachusetts Institute of Technology (MIT), Cambridge, in 2004. In 2004 he was a Postdoctoral Associate at the Laboratory for Information and Decision Systems, MIT. Currently, he is Assistant Professor at the Department of Electrical and Computer Engineering, University of Maryland, College Park. He is also affiliated with the Institute for Systems Research and his research interests include fundamental limits of feedback and the fusion between control theory and information theory.

 

Friday, April  7, 2006

3:30 – 4:30 p.m.

 1500 EECS