Contradicting a myth: good turbo codes with large memory order

Abstract

The purpose of this paper is to contradict a common myth about turbo codes. We are specifically addressing R=1/3 parallel-concatenated codes using systematic, recursive constituent codes. Myth: Turbo codes consisting of constituent codes with large memory order (i.e., a large number of trellis states) are not as effective as the original Berrou-Glavieux-Thitimajshima turbo code (1993) in the waterfall region of small signal-to-noise ratios (SNRs). This myth is contradicted by a turbo code whose recursive constituent codes have 256 states. Decoding this turbo code with the BCJR APP decoder gives bit-error-rate (BER) and frame-error-rate (FER) performance better than the original (Berrou) turbocode at all SNRs.