A performance analysis of LDPC code decoding under finite-precision arithmetics

While several implementations of LDPC codes decoders are available, the effects of fixed-point quantisation are still somewhat obscure. This paper aims at defining a systematic and general methodology to attack the LDPC decoding problem under finite–precision arithmetic. Statistical analysis is completed with Monte Carlo simulations to deploy the basics of the quantisation problem; the key aspect of this work is the independent study of channel reliability and extrinsic messages. As a case example, a short block, structured LDPC code is considered, and the results are shown. The extension to different codes confirms the validity of our method in relating design parameters and performance. This study is at the basis of the exploration of the trade-off between system performance and hardware complexity and can be easily extended to other applications