Speaker
Prof. Phil Schniter
Dept. of Electrical and Computer Engineering
The Ohio State University, Columbus OH, USA
This e-mail address is being protected from spambots. You need JavaScript enabled to view it
http://www.ece.osu.edu/~schniter

Date: Tuesday, July 8, 2008
Time: 1:00 – 4:00 pm
Location: Main NU Building (B2) – Auditorium.

Abstract
Channel variability (i.e., “fading”) is typically regarded as the principle challenge in wireless communication. For transmitters which operate without knowledge of the channel state, the data rate must be chosen low enough to maintain an acceptable error rate under a wide range of channel conditions, including poor ones, though doing so implies that the link will be under-utilized whenever the channel quality is high. This inefficiency motivates the design of adaptive transmission systems, whereby the data rate is optimized according to the transmitter’s knowledge of the channel state. In such systems, however, providing transmitter channel state information without compromising other system resources (e.g., power, bandwidth, and complexity) can be a serious challenge. Consider, for example, the difficulties involved in providing up-to-date channel estimates through a dedicated reverse link.
In this talk, we are motivated by a cross-layer approach to adaptive transmission whereby the physical layer uses feedback that is provided “for free” by the link layer, such as the packet-level ACK/NAKs used for ARQ, to adapt the data rate. Though not direct indicators of the channel quality, these ACK/NAKs can be considered as “relative” indicators of the channel quality, i.e., relative to the previously employed data rate. More precisely, we consider the problem of goodput-maximizing rate adaptation based on degraded causal error-rate feedback. By “goodput,” we mean the amount of data communicated without error. Because we employ relative (versus absolute) channel-quality feedback, the choice of data rate affects not only the subsequent goodput but also the subsequent feedback, implying a tradeoff between exploration and exploitation. This leads to a difficult stochastic optimization problem, in particular, a partially observable Markov decision process (POMDP) whose complexity and memory grow exponentially in time. For these reasons, we consider suboptimal rate schedules, and focus in particular on greedy rate adaptation. We then detail an implementation of our greedy scheme that allows the use of a continuous (versus finite-state) Markov channel model, and propose an even simpler scheme which adapts the transmission rate only once per block of packets. A numerical investigation of our greedy schemes shows that they achieve steady-state goodputs that are close to an upper bound on optimal goodput yet far from those of the best fixed-rate scheme. Furthermore, numerical investigations show that the drop rate induced by the use of a finite transmission buffer is also much better than that of the fixed-rate scheme.

Biography
Philip Schniter received the B.S. and M.S. degrees in Electrical and Computer Engineering from the University of Illinois at Urbana-Champaign in 1992 and 1993, respectively. From 1993 to 1996 he was employed by Tektronix Inc. in Beaverton, OR as a systems engineer. In 2000, he received the Ph.D. degree in Electrical Engineering from Cornell University in Ithaca, NY. Subsequently, he joined the Department of Electrical and Computer Engineering at The Ohio State University in Columbus, OH, where he is now an Associate Professor and a member of the Information Processing Systems (IPS) Lab.
Dr. Schniter’s areas of research include signal processing, communication theory, information theory, wireless sensor networks, and underwater acoustic communication.