低复杂度LDPC码信息瓶颈量化译码器设计

Design of a low complexity information bottleneck quantization decoder for LDPC codes

近年来,信息瓶颈方法已经被成功应用于低密度校验码量化译码器的设计当中,所设计的量化译码器仅需4 bit量化位宽就能逼近双精度和积译码算法的性能。此外,信息瓶颈译码器只需处理无符号整数,且可使用查表来替代复杂的校验节点运算,适宜于硬件实现。然而,现有的低密度校验码信息瓶颈译码器所需的查表次数与节点度数的平方成正比,不利于在含有高度数节点的低密度校验码(如有限几何低密度校验码和高码率低密度校验码)中应用。为了解决这一问题,提出了一种基于前后向算法的低密度校验码信息瓶颈量化译码器设计方案。在所提方案中,基于前后向算法的节点运算可以分为3个步骤:前向查表,后向查表以及输出外信息。为了降低存储空间,可以精心地设计使得前向查表和后向查表复用同一套表格行。在输出外信息时,节点充分利用了前向查表与后向查表时产生的中间结果,有效地避免了消息的重复计算,使得译码所需要的查表次数与节点度数呈线性关系。仿真结果验证了所提低密度校验码信息瓶颈量化译码器设计方案的有效性。

Recently,the information bottleneck(IB)method has been successfully applied to the design of quantization decoders for LDPC codes.The resulting high performance quantization decoder can approach the performance of the floating-point SPA(Sum-Product Algorithm)decoder by using only 4 quantization bits.Moreover,the LDPC IB decoder deals only with unsigned integers and replaces complex check node operations by simple lookup tables,and thus is very suitable for practical implementations.However,the number of table lookups is proportional to the square of the node degrees,which is unfavorable for LDPC codes with large node degrees,such as the finite geometric LDPC codes and high-rate LDPC codes.To deal with this issue,an improved scheme for the design of IB decoders is proposed based on the forward-backward algorithm.In the proposed scheme,the node operations based on the forward-backward algorithm can be divided into three steps:forward table lookup pass,backward table lookup pass,and extrinsic information generation.To reduce the memory space,a careful design can be carried such that the forward and backward table look-up passes share the same set of lookup tables.When generating extrinsic information,nodes make full use of the intermediate messages generated by forward and backward table lookups,and effectively remove redundant calculations,thus resulting in a total number of table lookups linear with node degrees.Numerical results are provided to demonstrate the effectiveness of the proposed LDPC IB decoder.