基于DenseNet的相关噪声抑制辅助译码算法OA

In practical mobile communication systems,correlated noise can significantly degrade the performance of belief propagation(BP)decoding algorithms.To address this issue,a channel denoising-aided decoding algorithm based on a dense convolutional network(DenseNet)is proposed.Without changing the basic framework of BP decoding,a plug-and-play DenseNet denoising module is introduced to enhance feature propagation and fusion through dense connections,thereby learning the correlated noise structure more effectively.The system adopts a cascaded structure of BP prior decoding,DenseNet residual denoising,and BP iterative decoding.A composite loss function combining mean square error and normality constraints is employed to reduce noise power while preserving the Gaussian property of residual noise.Simulation results based on low-density parity-check codes show that,when the noise correlation coefficient is 0.8,the proposed scheme achieves approximately 3 dB performance gain over conventional BP decoding without denoising and a further gain of about 0.5 dB over conventional convolutional neural network-aided schemes,verifying its effectiveness and application potential in correlated noise channels.