ABSTRACT
Until 1993 the most powerful error-correcting codes achieved a performance of about 2 dB from the theoretical limit. The difficulty in approaching this limit resides in the fact that as the codes become more powerful, their complexity increases and the respective maximum likelihood decoder (MLD) becomes computationally intractable. A breakthrough occurred with the development of turbo codes, which made it possible to significantly reduce the gap to the theoretical limit. These codes, introduced in Berrou et al. [1993], were shown to achieve near-Shannon limit performance on additive white Gaussian noise (AWGN) and Rayleigh flat fading channels. Although the maximum likelihood decoder for these codes has a prohibitively high complexity, its structure allows the implementation of a suboptimal iterative decoder.
