Slowest Descent Sequence Detection

Dr. Predrag Spasojevic

Dept. of Electrical and Computer Engineering
Rutgers - The State University of New Jersey
Piscataway, NJ

Thursday, April 12th, 11:00 AM, ENS 637

spasojev@winlab.rutgers.edu

Abstract

A novel approach to reduced complexity sub-optimal sequence detection termed the slowest descent method is proposed. It is closely related to the zero-forcing equalizer for inter-symbol interference channels and the decorrelator detector for the multi-user detection problem. Later two methods select the sequence which is closest to the unconstrained maximizer of the quadratic likelihood fonction. Thus, they can be ciewed as members of a class of generalized decorrelators which "quantize" a convex set maximizer of an (not necessarily quadratic) objective function. The slowest descent method compares the goodness of the generalized decorrelator estimate to a set of admissible sequences obtained based on an analysis of the objective function's slope in a neighborhood of the convex set maximizer.

Solutions based on the slowest descent search and the generalized decorrelator are studied for several sequence estimation problems including detection for coherent and non-coherent channels with multiple users, quasi-static multipath fading, and/or non-Gaussian noise.

The results stress the importance of the direction os the least decrease in the objective function for detection especially when the "effective channel" has zeros in its frequency response. When the slope in a neighborhood of a convex maximizer is evenly distributed in all directions, the generalized decorrelator estimate can have a close-to-optimal performance.

A list of Telecommunications and Signal Processing Seminars is available at from the ECE department Web pages under "Seminars". The Web address for the Telecommunications and Signal Processing Seminars is http://anchovy.ece.utexas.edu/seminars