This paper presents a stage-reduced odd-even sorter architecture for deeply pipelined successive cancellation list (SCL) polar decoders. First, by exploiting the relationships between the metrics of candidates maintained during the decoding process, u...
This paper presents a stage-reduced odd-even sorter architecture for deeply pipelined successive cancellation list (SCL) polar decoders. First, by exploiting the relationships between the metrics of candidates maintained during the decoding process, unnecessary compare-and-swap units (CASUs) are pruned. Subsequently, the CASUs are relocated to minimize the number of CASU stages to go through from the beginning to the end of the sorter. Moreover, to verify the efficacy of the proposed sorter architecture for arbitrary list sizes, generalized formulae are derived. As a result, compared with the original odd-even sorter, about 60% CASUs and a few stages can be removed.