LDPC decoding using multiple representations
ABSTRACT For a particular block code, a multitude of different parity check matrices could be chosen to represent the code constraints. We demonstrate that belief propagation decoders based on different parity check matrices often respond in different ways, and that for a given complexity, it may be preferable to use several decoders for fewer iterations, than to run a single decoder for many iterations. We also propose some additional decoder architectures that use multiple representations of the parity constraints.
[show abstract] [hide abstract]
ABSTRACT: We propose the application of multiple-bases belief-propagation, an optimized iterative decoding method, to a set of rate-1/2 LDPC codes from the IEEE 802.16e WiMAX standard. The presented approach allows for improved decoding performance when signaling over the AWGN channel. As all required operations for this method can be run in parallel, the decoding delay of this method and standard belief-propagation decoding are equal. The obtained results are compared to the performance of LDPC codes optimized with the progressive edge-growth algorithm and to bounds from information theory. It will be shown that the discussed method mitigates the gap to the well-known random coding bound by about 20 percent.10/2008;
Conference Proceeding: Improved iterative decoding of LDPC codes from the IEEE WiMAX standard[show abstract] [hide abstract]
ABSTRACT: Multiple-bases belief-propagation is a parallel decoding setup which allows for improved decoding performance when compared to standard belief-propagation. Originally designed for decoding of high-density parity-check codes in an iterative manner, this method also shows good decoding results for well-designed low-density parity-check codes when signaling over the AWGN channel. We show the applicability of this scheme to channel codes defined in the IEEE WiMAX standard. It is challenging to find sets of well-performing parity-check matrices for these codes, all of them differing from each other. We propose an algorithm which makes use of the special structure of an underlying base matrix to accomplish this task. The results are compared to codes constructed by the progressive edge-growth algorithm and to bounds from information theory.Source and Channel Coding (SCC), 2010 International ITG Conference on; 02/2010