Performance Balanced General Decoder Design for QC-LDPC Codes Using Vivado HLS
Bingbing Wang, Jing Kang, Yan Zhu · 2021
Field-programmable gate array are frequently used to implement high-speed Low-density Parity-check (LDPC) decoders. The conventional practice is to develop a register transfer level (RTL) description of a digital circuit, which requires considerable simulation and verification, resulting in a long development cycle. High-level synthesis (HLS) tools significantly accelerate the hardware design process by using software programming languages such as C/C++. Well optimized code can be synthesized by the HLS tools into efficient hardware circuits. In this paper, we designed a Xilinx FPGA-based LDPC decoder using Vivado HLS. The decoder takes advantage of the cyclic shift of the submatrix in the parity check matrix of the quasi-cyclic LDPC (QC-LDPC) codes and uses a wide-pipeline architecture to process one row/column of all non-zero submatrices per clock cycle, with decoding throughput of hundreds of Mbps. By using different arrays to store data on multiple diagonals in a submatrix can easily avoid the performance bottleneck caused by access conflicts. In addition, the decoder has strong generality and flexibility, which has a high practical value.