摘要
近年来,无线通信中对于传输延迟的要求越来越高。降低信道编码的码长,可以在信息速率较低的情况下,显著降低传输延迟。低密度奇偶校验码(LDPC)是当前广泛采用的信道编码,在LDPC的码长较短时,由于Tanner图上的环较多,周长小,因此传统的置信度传播算法对符号似然比的近似较差,从而影响了编码的性能。提出了基于循环神经网络的BP-RNN译码器,采用机器学习中的工具优化Trellis图中边的权重,与传统BP译码器相比,提高了短码长LDPC的译码器性能,从而在低信息速率的传输中降低通信延迟。
Recently there is a strong need for low transmission delay in wireless communications.When the information rate is low,it is a good practice to reduce the block length of channel codes.While LDPC is the most widely used codes currently,when the code length of LDPC is short,there are many circles in Tanner graph and the girth of Tanner graph is small.These small circles produced by the traditional belief propagation algorithm lead to the poor estimation of symbol likelihood ratio,and undermine the performance of the code.A BP-RNN decoder based on recurrent neural network is proposed and the tools in machine learning optimize the weights in Trellis graph.Simulation results show that RNN based decoder outperforms the traditional BP decoder for short-length LDPC.Thus the transmission delay can be significantly reduced in low information rate communication without sacrificing the performance of communication.
作者
倪永婧
郭巍
张静涛
NI Yongjing;GUO Wei;ZHANG Jingtao(College of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050000,China;The 54th Research Institute of CETC,Shijiazhuang 050081,China)
出处
《无线电工程》
北大核心
2021年第7期557-562,共6页
Radio Engineering
基金
国家部委基金资助项目。
