High-Throughput Adaptive List Decoding Architecture for Polar Codes on GPU
Zhanxian Liu, Rongke Liu, Haijun Zhang
Abstract
Polar codes have been adopted in the 5th generation communication standard (5G). Successive-cancellation list (SCL) decoding, a promising solution to decode polar codes, provides good tradeoff between error-correction performance and implementation complexity. To meet high flexibility and scalability requirements of Cloud or Open random access networks (RANs) for the 5G or future communication standards, this paper presents software list decoding architectures on graphics processing units (GPUs) and considers special nodes to reduce the decoding complexity. In order to improve the throughput and latency performance, a GPU-based adaptive mapping strategy taking advantage of the GPU architecture is proposed. Without considering the data transfer procedure between host and device, the proposed adaptive list decoder on NVIDIA RTX3090 for the code (1024, 512) with list sizes <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$L=8$</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$L=32$</tex-math></inline-formula> can achieve above 30 Gbps coded throughput at the frame error rates of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$10^{-6}$</tex-math></inline-formula> and <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$10^{-7}$</tex-math></inline-formula> , respectively. Compared to the state-of-the-art software works, the proposed decoder achieves about <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\times 1.78$</tex-math></inline-formula> to <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><tex-math notation="LaTeX">$\times 9.07$</tex-math></inline-formula> throughput speedups.