An FPGA Implementation of Two-Input LUT Based Information Bottleneck LDPC Decoders

Abstract

A lookup table-based check and variable node are considered for designing low-density parity check (LDPC) decoder architectures, using the principle of the information bottleneck method. It has been shown that an information bottleneck LUT operation can outperform conventional minsum arithmetic operation in terms of error-correction capability. This paper presents a cost-efficient hardware implementation of LUT-based node processing units in the decoder architecture. It exploits the symmetry of the communication channel and multi-input LUT decomposition to generate a reduced-size LUT structure. The LUT operations are designed as a two-level memory subsystem enabling LUT mappings reconfiguration at runtime. As a case study, a rate-7/10 7650-bit regular QC-LDPC decoder is implemented on an FPGA, achieving a throughput of up to 1.345 Gbps at 10 iterations. Compared with the conventional offset min-sum decoder, the proposed decoder increases the throughput-to-area ratio up to 39.22% at a cost of no more than 0.08 dB decoding performance loss. In addition, the hardware complexities of node design variants are investigated.