针对极化码置信度传播算法的低复杂度早期停止准则  

Low-complexity Early Stopping Criterion for Belief Propagation Decoding of Polar Codes

作　　者：张小军 李娜 董雁飞 崔建明 郭华 ZHANG Xiaojun;LI Na;DONG Yanfei;CUI Jianming;GUO Hua(College of Electronic and Information Engineering,Shandong University of Science and Technology,Qingdao 266590,China;State Key Laboratory of High-end Server and Storage Technology,Jinan 250101,China)

机构地区：[1]山东科技大学电子信息工程学院,青岛266590 [2]高效能服务器和存储技术国家重点实验室,济南250101

出　　处：《电子与信息学报》2021年第1期77-84,共8页Journal of Electronics & Information Technology

基　　金：山东省自然科学基金联合基金(ZR2019LZH001);山东省重点研发计划(2019GGX101066);山东省高等学校青创科技计划(2019KJN020,2019KJN024);泰山学者计划。

摘　　要：针对极化码译码延迟较高的问题,该文提出了一种针对置信度传播算法的早期停止准则,通过监测码字估值x的收敛性来终止译码。该准则利用高斯近似分析选取码字中Q个出错概率较小的比特构成比较空间,由于比较的位数较少,且仅采用异或和或运算,其计算复杂度较低。与基于信息序列估值u的方案不同,提出的准则在计算u之前已完成检测,不会导致额外的译码延迟。仿真和FPGA综合结果表明:该准则相对于G-Matrix,最坏信息位(WIB)和冻结位误码率(FBER)可有效节省硬件资源;当最大迭代次数设置为40次时,相比于G-Matrix准则,复杂度下降的代价是平均迭代次数在3.5 dB处上升了29.98%,相比于WIB和FBER方案,平均迭代次数分别减少39.44%和27.67%。Considering the high decoding latency of polar code, an early stopping criterion for belief propagation is presented, which terminates the decoding by monitoring the convergence of codeword estimate x. In this paper, Gaussian approximation is used to analyze and select Q bit with low error probability to construct the comparison space. Because the number of bit to be compared is small and only XOR and OR operation is used,the computational complexity is low. Different from other criteria based on u, the proposed criterion does not lead to additional latency for it has been completed before calculating u. Simulation and FPGA Synthesis results show that compared with G-matrix, Worst Information Bit(WIB) and Frozen Bit Error Rate(FBER),this criterion can effectively save hardware resource.When the maximum iteration number is set to 40,compared with the G-matrix criterion, the average iteration time is increased by 29.98% at 3.5 dB, and the average iteration times are reduced by 39.44% and 27.67% respectively compared with the WIB and FBER schemes.

关 键 词：极化码 置信度传播 早期停止准则 低复杂度 码字估值 

分 类 号：TN911.22[电子电信—通信与信息系统]