The
Conference
The International Symposium on Information Theory, ISIT 2001, was held at the
Omni Shoreham Hotel in Washington, D.C., USA, June 24-29, 2001.
The ISIT is the most recognized conference in the area of
information theory. It is a very theoretical conference, so the only companies
represented are book publishers.
Every morning starts with a one-hour plenary session
intended for a broader audience. This is then followed by six parallel
sessions, where typically two or three are within your own area. Each session
includes four technical talks limited to 20 minutes. There are two sessions
before lunch and two after, resulting in a total of 345 technical papers. Each
talk is accompanied by a one-page abstract in the conference proceedings.
However, the papers that appear in the proceedings have been selected by
peer-review based on a six page extended abstract.
Even though this is a very big conference, it is still fairly easy to get an overview of new things happening within your own area. The conference is like a big newsletter with close to everybody participating, either by presenting a paper or just by attending the conference.
Usually if you
find an interesting abstract in the proceedings you would attend the
presentation, which then gives a little more information. If you find the
presentation interesting you usually contact the presenter off line to ask for
the original six-page manuscript they sent for peer-review.
My
impressions
The purpose of attending the conference was to get new
ideas and a general understanding of the ongoing research in this area. My own
research project is a combination of information theory and real-time
communication. Since I have participated in some real-time conferences before
(ECRTS 1999, ECRTS 2000 and RTCSA 2000) it was very interesting to participate
in a pure communication conference this time.
A very hot topic is iterative decoding, i.e., a sub optimal
decoding method with good accuracy and significantly lower complexity.
Generally it is known that most block codes are good, provided that they are
long enough. However, the complexity of optimal decoding is NP-hard and
exponentially increasing with code length. Recently, a sub-optimal coding and
decoding strategy has been discovered, where the output of the sub-optimal
decoder is feed back again in order to be refined iteratively. This procedure,
termed iterative decoding or turbo-decoding, has performance close to optimal
with manageable decoding complexity.
For real-time purposes turbo-decoding is a very interesting
topic. We want to benefit from the high quality that the long codes provide,
but at the same time keep the decoding complexity, and hence the time to decode,
at a minimum. Obviously, the convergence behaviour is of utmost importance,
since a slow or no convergence reduces the performance, as does convergence to
an incorrect solution, i.e. errors. Therefore, one paper in particular was very
interesting to me:
Convergence and errors in turbo-decoding by Andrew Reid, University of Canterbury, Aaron Gulliver, University of Victoria and Desmond Taylor, University of Canterbury.
Here, the convergence for different turbo-codes was plotted as a function of the number of iterations, which is a measure equivalent to time.
This paper has been very useful indeed to me and is referenced several times in my licentiate thesis. Also, I have plotted the convergence versus the number of iterations for my particular decoder.
Twelve different sessions concerned iterative decoding and related topics so it is indeed a very hot topic.
Conclusions
The ISIT gives an excellent research overview, since basically everybody involved or interested in information theory attends the conference. There are also some recent work sessions and good social activities further improving the possibilities to get to know the different research groups. Another reason for the high participation is that within the IEEE Information Theory Society there is only one conference (ISIT) and one workshop (ITW). My overall opinion of the conference could not be better.