Multiple-Bases Belief-Propagation with Leaking for Decoding of Moderate-Length Block Codes

Inhalt:
Short algebraic codes promise low-delay data transmission and good performance results when transmitted over the additive white Gaussian noise (AWGN) channel and decoded by maximum-likelihood (ML) soft-decision decoding. One reason for this is the large minimum distance of these codes. For belief-propagation (BP) decoding, short algebraic codes show suboptimal results due to their high-density parity-check matrices. Using a collection of parity-check matrices for a given code, Multiple-Bases Belief-Propagation (MBBP) allows for decoding of many dense linear block codes with near ML performance. One convenient way to generate these matrices is using the automorphism group of a code. We point out limits of this approach and show two novel improvements, a decoder modification and a construction algorithm for parity-check matrices, which emphasize that MBBP is a more general approach and independent of the automorphism group. We use these methods to extend the field of application for MBBP to codes of moderate length. This includes codes with parity-check matrices tailored for BP decoding, in particular Progressive Edge-Growth (PEG) codes.