New Achievable Rates for the Multiple Access Channel with Feedback

It is a well-known result of Shannon that feedback does not increase the capacity of a point-to-point memoryless channel. In contrast, the capacity region of a memoryless multi-user channel may be enlarged using feedback. This was first demonstrated by Gaarder and Wolf in 1975 through an example with a binary erasure multiple-access channel (MAC). Building on the example, Cover and Leung established an achievable rate region for the MAC with feedback. However, the existence of a 'single-letter' characterization for the MAC feedback capacity remains an open question.
In the talk, we will develop the concept of “typicality graphs” and show how the Gaarder-Wolf and Cover-Leung coding schemes can be interpreted using this framework. This interpretation suggests a natural generalization of the Cover-Leung scheme to yield new achievable rates. We will show that this rate region is strictly larger than the Cover-Leung region using the white Gaussian MAC as an example. Comparisons with other coding schemes for the MAC with feedback will also be discussed.

Ramji Venkataramanan received his M.S and Ph.D in EE: Systems from the University of Michigan, Ann Arbor. Since October 2008 he has been a post-doc at the Information Systems Lab at Stanford, working on problems in stochastic network theory. His research interests include information theory, networking and any interesting applications of probability.