Delay, Feedback, and Interaction

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

Information theory is the strategic theory of communication: providing architectural guidance and fundamental bounds. For communication, there are three core quality of service parameters: probability of error, rate, and end-to-end delay. Shannon's seminal capacity theorems establish bounds on rate as the tolerated probability of error goes to zero while the acceptable end-to-end delay goes to infinity. However, for many applications like telemedicine, remote control (e.g. fly-by-wireless in UAVs), and even video-conferencing, short delays are also critical. Unfortunately, it turns out that the classical block-code oriented approaches to information theory are misleading regarding the tradeoffs involving end-to-end delay, especially when feedback is involved. This research aims to remedy that situation and thereby get a deeper understanding of the nature of information flows and their communication requirements. Having these fundamental architectural results will help guide not only the design of next generation communication systems, but is also critical for the setting of regulatory policy governing wireless spectrum since both interactive and non-interactive applications must coexist efficiently in the wireless context. This research makes three core contributions: (a) Characterizing the tradeoff between delay and probability of error in communication systems with and without feedback in the context of both soft and hard latency constraints by leveraging our new hallucination bound and uncertainty focusing bound techniques, (b) Understanding how to exploit noisy and limited feedback in the above contexts, (c) Deepening our understanding of information flows in the context of interactive control and distributed estimation by taking a new approach to source-channel separation theorems based on the theoretical CS model of showing problem-level equivalences through explicit reductions of one communication problem to another.