BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Topics on joint source-channel coding and multiuser detection - Ad
 ria Tauste Campo\, Signal Processing and Communications Laboratory\, CUED
DTSTART:20110705T131500Z
DTEND:20110705T140000Z
UID:TALK32000@talks.cam.ac.uk
CONTACT:Rachel Fogg
DESCRIPTION:This talk summarises my doctoral research over the last 3 year
 s.\n\n\nIn real-time applications\, data blocklength/delay is a design con
 straint of many point-to-point communication systems \nmodeled by joint or
  separate source-channel coding blocks.  Information theory traditionally 
 studies such systems \nin the asymptotic regime and concludes that there i
 s no loss in optimality \nby using separate instead of  joint source-chann
 el coding when the delay/blocklength goes to infinity.  This is why due to
  the simplicity of a separation design source and channel coding are indep
 endently implemented in most applications. \n\nHowever\, this is no longer
  true when packets of hundreds of bits are sent over a noisy channel\n(e.g
 . wireless systems). In this situation joint coding schemes are expected t
 o beat separation in terms of error probability\nbut there is little knowl
 edge on the actual gap. We characterize this gap by first providing new ac
 hievability and converse error bounds for both joint and separate source-c
 hannel coding and second by refining previous results on the  error expone
 nt. \nOne could expect that  the gap is optimized by fully exploiting  the
  dependence of the channel input onto the source messages. However\, our  
 results show  that this dependence is optimally weak in most of the cases.
  \n\n\nAnalysis and design of optimum (Bayes) detection of users' (mobiles
 \, tablettes\, etc.) communicating with a base station \nis in general pro
 hibitive. However\,  when we assume that the number of such users is large
 \, there are \nstatistical-physics techniques that fairly characterize the
  performance of finite systems (e.g.\, down to 8-10 users).  That perfoman
 ce\nis typically specified by a fixed-point equation with multiple solutio
 ns\, but  only a subregion of those is of interest for practical purposes.
   We provide a formula for this subregion and derive conditions for near-o
 ptimal performance in terms of the number of users\, their activity\, and 
 their power.  We also study a scheme where both data and activity are enco
 ded and propose an iterative algorithm to jointly detect users' data and a
 ctivity in the asymptotic regime. \n
LOCATION:LR5\, Engineering\, Department of
END:VEVENT
END:VCALENDAR