High-performance software implementations of SCAN decoder for polar codes

摘要

This paper presents the first optimized software implementation of a SCAN decoder for Polar codes. Unlike SC and SC-List decoding algorithms, the SCAN decoding algorithm provides soft outputs (useful for, e.g., parallel concatenated decoders Zhang et al. IEEE Trans Commun 64(2):456–466 2016). Despite the strong data dependencies in the SCAN decoding, two highly parallel software implementations are devised for x86 processor target. Different parallelization strategies, algorithmic improvements, and source code optimizations were applied in order to enhance the throughput of the decoders. The impact of the parallelization approach, the code rate, and the code length on the throughput and the latency is investigated. Extensive experimentations demonstrate that the proposed software polar decoder can exceed 600 Mb/s on a single core and reaches multi-Gb/s when using four cores simultaneously. These decoders can then achieve real-time performance required in many applications such as software defined radio or cloud-RAN systems where network physical layer is implemented in software.