From Universal Channel Coding to the Tracking of Stopping Times

SPEAKER: Dr. Aslan Tchamkerten, Postdoctoral Associate, MITABSTRACT: We consider point-to-point communication over discrete memoryless channels. Since early 1960's it has been known that, without feedback, the set of achievable rates strongly depends on whether the channel statistics are revealed to the communicating parties. In contrast, if noiseless feedback is available, there is no rate loss under very general assumptions. However, if we now consider the second order question "what is the optimal tradeoff between delay and error probability for universal communication?" the answer is far from clear. Are there blind feedback schemes that perform as quickly and as reliably as if the channel were revealed to the communicating parties? Except for trivial cases (e.g., binary erasure channel) this has been an open question since mid 1970's.In the first part of the talk, we first show that there exist non-trivial families of channels for which universally optimal feedback schemes exist. Maybe somewhat surprisingly, these schemes are not training based. Second, given a pair of channels, we give a simple criterion under which no universally optimal scheme exist for that pair.As part of our quest for robust communication, in the second part of the talk, we address the situation where the feedback channel is noisy. This setting leads to an important synchronization issue when encoder and decoder decide on the basis of stopping times. We provide insights into this issue by solving a new statistical problem, the tracking stopping times problem. Interestingly, this problem is a generalization of the celebrated (Bayesian) change-point problem, and it has several applications in other areas, such as detection and forecasting.This is based on joint work with U. Niesen, E. Telatar, and G. Wornell.Bio: Aslan Tchamkerten received the engineer physicist diploma from the Ecole Polytechnique Federale de Lausanne (EPFL), Switzerland, in 2000. After finishing the graduate school in communication systems at EPFL in 2001, he started his Ph.D. in the Information Theory Lab., and graduated in 2005. Since 2005 he has been a postdoctoral associate at MIT (EECS).Host: Prof. Giuseppe Caire, caire@usc.edu

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Mayumi Thrasher