A CNN-Aided Post-Processing Scheme for Channel Decoding Under Correlated Noise
Junhao Liu, Shaoli Kang, Jing Cheng, Jiaqing Wang, Ruihua Huang · 2024
In this paper, a convolutional neural network-aided post-processing scheme (CNNAPS) is proposed for improving the channel decoding performance of the denoiser under correlated noise. In this scheme, the CNN estimates actual channel noise for computing the error pattern, and then error nodes are identified by a threshold discrimination method. Specifically, the channel log-likelihood ratio (LLR) value of the identified error node is set to zero, while the received signal of the identified correct node is denoised. Furthermore, one of the key dvantages of CNNAPS is that it does not require knowledge of code structure properties, making it applicable to all linear block codes. Simulation results demonstrate that CNNAPS effectively identifies error nodes within error blocks. It is also shown that by implementing the post-processing along with a single redecoding, CNNAPS can significantly improve the error correction performance of the denoiser with minimal increase in complexity.