Design of Min-Sum-based LDPC decoders using imprecise arithmetic

This work evaluates the robustness of Low-Density Parity-Check decoders against errors due to imprecise arithmetic. While the use of imprecise arithmetic is motivated by savings in energy, delay and area, it also causes errors during the decoding process. This is a new paradigm in coding theory, which traditionally assumes that an error correction decoder operates on exact hardware and errors can only be introduced by the transmission channel. We design imprecise arithmetic operators and investigate their use within several Min-Sum-based decoders. We show that all decoders are able to provide error protection, but most of them suffer a performance penalty compared to the exact arithmetic implementation. Remarkably, the Self-Corrected Min-Sum decoder incurs no performance penalty when imprecise arithmetic is used.