-
Thoughts on Mixing for EXIT Charts
Thu, Apr 05, 2007 @ 03:30 PM - 04:30 PM
Ming Hsieh Department of Electrical and Computer Engineering
Conferences, Lectures, & Seminars
SPEAKER: Dr. Jossy Sayir, Telecommunications Research Center (FTW)ABSTRACT: EXIT charts and density evolution are two analysis techniques
to predict the peformance of iterative algorithms. EXIT charts
track the evolution of the average bitwise mutual information,
while density evolution tracks the evolution of the average
message distributions. In this talk, I will investigate the
difference between averaging mutual informations and averaging
distributions. A recent paper by Lechner, Kramer and Pedersen
shows that for messages that satisfy the symmetry condition
(e.g., Log-Likelihood Ratios), averaging distributions or
mutual informations are equivalent. For sub-optimal algorithms
(e.g., min-sum, Gallager A&B, etc.), where messages do not satisfy
the symmetry condition, they suggest a modification of EXIT chart
analysis that averages distributions in order to get the correct
prediction. We show that this does not necessarily yield the
correct prediction in all cases, and propose an alternative method
based on averaging mutual informations that does.This is joint work with Ingmar Land.BIO: Dr. Jossy Sayir received his Dipl. El.-Ing. degree from the ETH Zurich
in 1991. From 1991 to 1993, he worked as a development engineer for
Motorola Communications in Tel Aviv, Israel, contributing to the
design of a digital mobile radio system. He returned to ETH from 1993
to 1999, getting his PhD in 1999 under the supervision of Prof. J.L. Massey.
His thesis "On Coding by Probability Transformation" covered topics in
data compression, joint source-channel coding using arithmetic
codes, and capacity computation algorithms. Since 2000, he
has been employed at the Telecommunications Research Center (FTW) in
Vienna, Austria, as a senior researcher. His research interests
include iterative decoding methods, sub-optimal and quantized decoders,
and wireless ad hoc and sensor networks. Since July 2002, he manages
part of the strategic research activities at FTW and supervises a group
of researchers. He has taught courses on Turbo and related codes at
Vienna University of Technology, at the University of Aalborg,
Denmark, and at the University of Adelaide, Australia. He has served on
the organization committees of several international conferences and
workshops. In his spare time, he loves to cook, and plays the alto saxophone
in various jazz ensembles.HOST: Giuseppe Caire, caire@usc.edu
Location: Hughes Aircraft Electrical Engineering Center (EEB) - -248
Audiences: Everyone Is Invited
Contact: Mayumi Thrasher