International Symposium on Computing and Networking Workshops (CANDARW)
GPU implementation of binary and non-binary LDPC decoding
Mark Mizzi, Johann A. Briffa
Abstract
Since their rediscovery in the 1990s, Low-Density Parity-Check (LDPC) codes have dominated the field of capacity-approaching error correcting codes, and are currently used in a number of practical applications. The recent use of these codes in quantum key distribution (QKD) systems often requires much larger codeword sizes than previously considered, and has renewed interest in LDPC codes with non-binary alphabets. Both aspects introduce a significant increase in computational complexity, so that real-time implementation in practical QKD systems remains a challenge. In this work we propose a GPU-accelerated LDPC decoder for binary and non-binary alphabets, evaluating its performance in comparison to a reference CPU implementation and to earlier GPU and CPU implementations, for a range of field sizes and codeword lengths.