基于GPU的并行Turbo乘积码译码器

Turbo乘积码是一类前向纠错码,在高码率下具有良好的误码率性能。TPC编码器的实现相对简单,其译码器的译码复杂度也比较合理。因此,TPC被广泛用于各种场景,例如卫星通信系统和数据存储系统等。提出了一种基于GPU的并行TPC译码器,可以同时译码二维乘积码矩阵的所有行或列。设计了一种并行基本译码器,以简化由扩展汉明码构成的TPC的译码过程。实现了测试样例和有效码字计算的并行化,降低了译码延迟。为了进一步提高译码吞吐率,提出了多通道TPC译码器。在不同的GPU上测量了并行译码器的性能,实验结果表明,与基于CPU的TPC译码器相比,基于GPU的并行TPC译码器的译码延迟显著降低。此外,基于GPU的并行TPC译码器的吞吐率在NVIDIA RTX 2080 Ti上达到30 Mbps,在NVIDIA GTX Titan V上达到38 Mbps,是基于CPU的TPC译码器性能的44倍和54倍。

A parallel Turbo product decoder on graphics processing unit

Turbo Product Code (TPC) is a class of Forward Error Correction (FEC) codes that have good Bit Error Rate (BER) performance at high code rate. TPC is widely used in a variety of scenarios, such as satellite communication systems and data storage systems. This paper proposes a GPU-based parallel TPC decoder. In it, all rows or columns of the two-dimensional product code matrix can be translated at the same time. A parallel basic decoder is designed to simplify the decoding process of TPC consisting of extended Hamming code. The parallelization of test sample and effective code word calculation is realized, and the decoding delay is reduced. In order to further improve the decoding throughput, we propose a multi-channel TPC decoder. In addition, the performance of parallel decoders is measured on different GPUs. The experimental results show that the decoding delay of the GPU-based parallel decoder is significantly reduced compared with the CPU-based TPC decoder. In addition, the throughput of the GPU decoder reaches 30 Mbps on the Nvidia RTX 2080 Ti and 38 Mbps on the NVIDIA GTX Titan V, which is 44 times and 54 times the performance of the CPU-based decoder.Key words: