.png){kind=icon}
.png){kind=icon}
Return
A Task-Guided Normalized Min-Sum Decoding Network for LDPC Codes-Based DJSCC
.png){kind=icon}
PRE
AI
DOI：10.1109/LCOMM.2023.3281576.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Distributed joint source-channel coding (DJSCC) refers to the correlated sources which are encoded separately and reconstructed in the centre node by joint decoding. Therefore, the challenging problem involves designing a simple and efficient decoding algorithm to utilize the correlations among distributed sources. In this letter, a multi-task deep learning decoder is proposed for the low-density parity-check (LDPC) code-based DJSCC system to improve the performance. This proposal is based on the shared neural normalized min-sum (SNNMS) decoding network and the log likelihood ratio (LLR) shared unit is added to further exploit the source correlation. Moreover, an adaptive multi-loss function is proposed to prevent the network tilting the balance in favor of one certain task. Simulation results indicate that the proposed decoder can achieve a significant performance improvement with almost the same complexity compared with the separated SNNMS decoders.
Keywords:
Distributed joint source-channel coding
low-density parity-check
min-sum
neural network