Zigangirov
Le Jeudi 11 Janvier 2001
Horaire exceptionnel : 10h30
au LRI, Salle 101
K. Zigangirov
(Department of Information Technology, Lund Institute of Technology,
Suède)
Asymptotical analysis of iterative decoding of turbo-codes
Résumé/Abstract :
This is a joint work by K. Sh. Zigangirov, M. Lentmaier and D. V.Truhachev.
We analyse a conventional rate 1/3 turbo-coder, consisting of two parallel
rate 1/2 systematic recursive convolutional encoders and an interleaver
(scrambler). Analysis consists of two parts. First, we study the problem of
optimal interleaver choice, i.e., construction of an interleaver whose graph
has maximal length of the minimal cycle. Particularly, we prove a
turbo-code analog of the Gallager-Margulis theorem. Second, we analyse the
iterative decoding process assuming that decoder operates on a cycle-free
graph. To get analytical results we introduce a two-phase algorithm; the
first phase is analysed numerically, the second analytically. We prove
existence of iterative limits, i.e., an infimum of signal-to-noise ratios,
for which bit error probability increases at least exponentially with the
number of iterations. We upperbound these limits for concrete constructions.