LDPC decoder with embedded coding on unreliable memories

Unreliable message storage severely degrades the performance of LDPC decoders. This paper discusses the various impacts of bit errors of finite-precision messages on LDPC decoders. Discrete density evolution indicates that the sign bits of finite-precision messages have the most influence on the decoding threshold. As a result, this paper proposes to protect the sign bits of messages by Hamming product codes. Simulation results show that the proposed scheme only has a degradation of 0.1 dB for the LDPC decoder with a storage error ratio of 0.001, which outperforms the traditional triple modular redundancy scheme.