A RECURRENT NEURAL NETWORK-BASED SUCCESSION CANCELLATION FOR POLAR DECODER

To solve the low output and high latency of successive cancellation decoding for polar codes in 5G scene, accelerate scheme of successive cancellation (SC) decoding based on recurrent neural network (RNN) for polar codes is proposed. In this work, we leverage the expert knowledge in communication systems and adopt a long short-term memory network (LSTM)-aided SC to improve the performance of conventional SC. To lower the complexity, we consider a method of pruning the children binary tree which leaf nodes are all frozen bits or information bits under the chosen different parameters, according to the reliability of the polarized different channels. Simulation shows that the proposed scheme has a better decoding performance compared with successive cancellation decoding, and also reduces the time complexity of the successive cancellation decoding based on deep learning.