Synchronization in Turbo Coded Systems
Turbo codes and their unusual properties have attracted significant attention
of the information theory community. There already exist many articles
and papers devoted to the theory of these codes. This thesis, however,
focuses on practical implementation issues of turbo codes, with synchronization
problems as the main topic.
Turbo codes are usually discussed under assumptions of perfect channel
estimation and synchronization. It simplifies the analysis but does not
show their true performance in realistic conditions. This thesis investigates,
among others, the effects of non-perfect timing and phase offset recovery
on bit error rate. It is shown that non-perfect synchronization can severely
impair turbo coding gains.
The typical ML synchronization algorithms may not be a good solution
for turbo code systems, mostly due to the low signal-to-noise region in
which such codes operate. This work postulates using the soft bit output
of the BCJR algorithm as the source of better statistics for synchronization
algorithms. Two algorithms using soft bits for timing recovery are presented
and shown to outperform the classical ML NDA method. The phase offset compensation
can also be significantly improved when using soft bits.
The salient advantage of using the BCJR algorithm for synchronization
purposes lies, however, in its ability to detect non-linear phenomena of
cycle-slip and M-fold phase ambiguity. This eliminates the necessity of
transmitting additional pilot symbols, which increases the available bandwidth.
maximum a-posteriori estimation
maximum likelihood estimation