基于时间卷积网络的极化码译码算法-《计算机技术与发展》

文章信息/Info

Title:: Polar Codes Decoding Algorithm Based on Temporal Convolutional Network

作者:: 李硕¹ ; 王友国^2* ; 柴允¹ ; 任珈仪²; 1. 南京邮电大学通信与信息工程学院,江苏南京 210003;
2. 南京邮电大学理学院,江苏南京 210023

Author(s):: LI Shuo1 ; WANG You-guo2* ; CHAI Yun1 ; REN Jia-yi2; 1. School of Telecommunications and Information Engineering,Nanjing University of Posts and Telecommunications, Nanjing 210003,China;
2. School of Science,Nanjing University of Posts and Telecommunications,Nanjing 210023,China

Keywords:: polar code; SC decoder; temporal convolutional network (TCN); dilated causal convolution; residual connections

摘要:: 针对传统的极化码译码(SC 译码) 算法实际应用中的用时较长和容错率较差的问题,提出并使用新型人工神经网络——时间卷积网络 ( temporal convolutional network, TCN)? 拟合的方式搭建极化码译码模型。与其他人工神经网络不同的是, 时间卷积网络属于卷积神经网络( convolutional neural network,CNN) ,和循环神经网络( recurrent neural network,RNN) 的功能相似,其独有的膨胀因果卷积结构和残差链接方法使其擅于分析时间数据,比长短期记忆网络( long short-term memory, LSTM) 、门控循环神经网络( gated recurrent units,GRU) 之类的规范循环网络更准确、更简单、更清晰,比较适合极化码这样的时间序列。通过调试网络模型参数的方式,对时间卷积网络译码性能影响进行了研究,仿真结果显示,通过合理地调整训练序列数、卷积核的大小和数目可以实现提升极化码译码性能的要求。

Abstract:: Aiming at the problems of long time and poor error tolerance in the practical application of traditional polar code decoding (SC decoding) algorithms,a new type of artificial neural network,temporal convolutional network (TCN),has been declared to build a polar code decoding model. Different from other artificial neural networks,TCN belongs to CNN,which is similar in function to RNN. It has aunique dilated causal convolutions structure and residual connections method,which is more accurate,simpler and clearer than canonical recurrent networks such as LSTM and GRU,which is more suitable for time series such as polar codes. By adjusting the parameters of the network model,the effect of time convolutional network decoding performance can be observed. The phenomena of simulations have proven that by reasonably adjusting the number of training sequences,the size and number of convolution kernels,the requirements of improving the polar codes decoding performance can be achieved.