Optimization and Improvement of BP Decoding Algorithm for Polar Codes Based on Deep Learning

Abstract In order to solve the high latency problem of polar codes belief propagation decoding algorithm in the 5G and the dimension limitation problem of belief propagation decoding algorithm under deep learning, a multilayer perceptron belief propagation decoding (MLP-BP) algorithm based on partitioning idea is proposed. In this work, polar codes is decoded using neural networks in partitioning, and the right transfer message value of BP decoding algorithm is also set to complete the propagation process. Simulation results show that, compared with BP decoding algorithm, the proposed algorithm has better decoding performance, reducing the decoding latency, and it is also applicable to long polar codes.