Systems Seminar

Exploiting Sparsity in Equalization Algorithms,
or
What's wrong with my HDTV?

Prof. Bill Sethares
UW ECE Dept.

Abstract

The US High Definition Television broadcast standard was set several years ago, but even the latest receivers can fail "inexplicably". Transmission channels may have dozens of multipath interferers, and these may be spread over (up to) 1000 symbols. Because the training sequences are short, blind equalizers (adaptive linear + adaptive DFE) with hundreds of taps are needed, even though most of the taps are zero most of the time. How can this *sparsity* be codified and exploited?

This talk reports on a series of experiments in which we measured HDTV transmissions, identified the channels, and built appropriate equalizers. We then borrowed some ideas from machine learning, which, with appropriate reformulation, show how to exploit the sparsity of the equalizers in an efficient way. The resulting algorithms, translated into practical (blind and adaptive) equalizers may have applications in other systems problems where a priori information is available.

Time and Place: Wed., Oct. 11, 3:30-4:30 pm in 4610 Engr. Hall.

SYSTEMS SEMINAR WEB PAGE: http://www.cae.wisc.edu/~gubner/seminar/