基于深度残差收缩网络的LDPC译码算法

LDPC decoding algorithm based on deep residual shrinkage network

为了研究瑞利衰落信道下提高低密度奇偶校验码(low density parity check,LDPC)信道译码算法纠错性能的方法,结合神经网络技术,提出一种基于深度残差收缩网络(deep residual shrinkage networks,DRSN)的归一化最小和(normalized min-sum,NMS)译码算法(简称DRSN-NMS译码算法)。首先,本译码算法使用深度残差收缩网络预测信道增益;然后结合接收信号计算对数似然比(log likelihood ratio,LLR),将其作为译码算法的输入进行译码,DRSN通过学习接收信号中噪声的相关特征,以抑制噪声的方法使预测结果更加接近真实信道增益;最后使用实现较简便的NMS算法进行译码。仿真试验结果表明,在高信噪比环境下,本译码算法的误码率最低时接近常规算法误码率的1/3,译码性能得到一定的提高。本研究结果可为译码算法降低误码率提供参考。

In order to explore how to improve the error correction performance of the LDPC channel decoding algorithm under the Rayleigh fading channel,a normalized minimum sum(NMS) algorithm was proposed on the basis of the deep residual shrinkage network(DRSN) decoder in combination with the knowledge of neural network.Firstly,the algorithm(abbreviated as DRSN-NMS) used the deep residual shrinkage network to predict the channel gain and then calculated the log likelihood ratio(LLR) as the input of the decoder by combining the received signal.By learning the related features of noise in the received signal to suppress the noise,the deep residual shrinkage network makes the predicted results closer to the real channel gain,and then conducted decoding with the simpler normalized minimum sum algorithm.Simulation results show that in high SNR environment the bit error rate of the proposed algorithm is at its minimum close to 1/3 of that of the conventional algorithm,with,the decoding performance improved to a certain extent.The results can provide reference for the decoding algorithm to reduce the bit error rate.