Stochastic Observability, Adaptation & Internet Congestion Control

Speaker: Professor Robert Bitmead 

Host Faculty: Professor Zhong-Ping Jiang

Abstract

Internet congestion control is a specialized feedback control problem where, in TCP/IP, the feedback consists of acknowledgement, or ACK, packets and is used by the source computer to regulate input data rates. This situation is recast as a state estimation problem, where the state is that of a bottleneck router and consists of buffer space and competing traffic state elements. The state evolution is described by a Hidden Markov Model. Observability of this state from the source is then studied and the usual definition of observability found lacking for such stochastic system. A new definition is proposed, which reduces to the usual one in linear deterministic systems. For Hidden Markov Models however, this new definition sheds new light on the role of the feedback control law in maintaining observability and on the control performance price paid for adaptation in general.

About the Speakers

Bob hails from Australia originally and was at the Australian National University for 17 years before coming to UCSD in 1999. He will officiate as a Field Umpire at the US National Championships of Australian Rules Football in Mason Ohio shortly before coming to PINYU. He is a Fellow of IEEE, of IFAC, and of the Australian Academy of Technological Sciences & Engineering. He works in Control Systems, Signal Processing and Telecommunications, with an interest in the areas of intersection. He is a regular consultant to industry in the areas of modeling, estimation, and control. He was Associate Vice-Chancellor for Academic Personnel at UCSD from 2006-2009 and brews his own beer.