Mudri: A Fault-Tolerant Decoding Algorithm

We propose an improved version of probabilistic gradient descent bit flipping algorithm for decoding low density parity check codes, based on MUltiple Decoding attempts and Random re-Initializations (MUDRI). The proposed algorithm significantly increases the probability of correcting error patterns uncorrectable by the existing variants of bit-flipping algorithm. The performance of the algorithm implemented in noisy hardware is analyzed for various code types and codeword lengths, and shown to be superior compared to other hard decision algorithms. The MUDRI decoder is mostly insensitive to the failures in registers and logic gates and therefore represents a desirable solution for implementation in unreliable hardware.