Adaptive package coding on unreliable memories for LDPC decoders in radiation environment

The commercial off-the-shelf technology, advanced by NASA, has driven the utilization of commercial components in aerospace, which results in severe fault problems for electronic systems. This paper considers the implementation of LDPC decoders on the unreliable memories, which introduce errors to the iterative messages. Firstly, LDPC decoders are typically implemented with the partially parallel architectures, where the messages are synchronously produced in groups. Based on this, we propose to put the messages into packages so as to permit an efficient coding on the package instead of the single message. Secondly, the sign bit of the message is shown to have the most influence on the decoding, while the least significant bit (LSB) is less important especially when the message has a large magnitude. Thus, by evaluating the magnitude level, we introduce an adaptive coding on the sign bits utilizing the LSBs. And we utilize a majority logic decision to achieve the evaluation, which makes the decoding execution more reliable. We develop a discrete density evolution to analyze the performances of various schemes, where the proposed scheme is shown to have a coding gain of 0.2 dB to the existing scheme, as well as a residual error that is nearly one order of magnitude lower. Moreover, the results of finite length codeword simulations are well matched with the density evolution analysis.