Dynamic multi-bit successive cancellation list decoding of polar codes

Abstract

Polar codes are the first capacity-achieving error-correcting codes (ECC) which have attracted much research attentions. Among decoding methods, successive cancellation list decoding (SCLD) is considered as a representative decoding method to achieve good error-correction performance for finite length polar codes. However, the long latency of the SCLD is a significant problem to use it practically, which is mainly caused by two features of SCLD process : tree traversing and path splitting. To solve the main two reasons, a dynamic multi-bit successive cancellation list decoding (DM-SCLD) algorithm is proposed which decodes several bits simultaneously without traversing the tree to the leaf nodes. Furthermore, it reduces the path splitting, which is originally required when decoding every single bit, by finding a specific condition based on LLR values of the decoding tree nodes. In addition, an ignoring path(IP) technique is proposed to mitigate the condition of DM-SCLD which results in additional latency reduction compared to DM-SCLD. The simulation result show that the DM-SCLD with IP reduces clock cycles by 77%, 79% compared to conventional SCLD for List size = 8 and 16, respectively, with negligible error-correction performance degradation (<0.05dB).