High Performance Error Correction Under Low SNR Based on Deep Neural Network

The signal-to-noise ratio of long-distance relay free communication systems and quantum secure communication systems is very low, making it difficult to achieve high-performance error correction. In order to achieve effective error correction, longer code words or higher iterations are generally chosen to achieve the highest error correction efficiency as possible. However, this will greatly increase the complexity of the error correction process and reduce real-time performance. To solve this problem, we propose a high-performance error correction algorithm based on deep neural network. In the case of low signal-to-noise ratio, the error correction performance of this method is better than that of the classical LLR-BP algorithm for high-speed short codes. At the same time, the number of hidden layers and neurons were adjusted according to the coding length, and the generalization performance of the model was improved. The results show that our proposed DNN based decoding algorithm can achieve better error correction performance than LLR-BP when the signal-to-noise ratio is less than 9dB.