Blind Identification of LDPC Code Based on Deep Learning

Abstract

In cognitive radio or military communications systems, the receiver usually needs to blindly identify which LDPC code has been adopted by the transmitter. Existing methods for blind LDPC code identification suffer from high computational complexity. This paper proposes a deep learning based LDPC code identification algorithm. According to the algorithm, the received LDPC encoded sequence is treated as a text sentence, and a special convolutional neural network (CNN), TextCNN, is utilized to understand the sequence and infer which code is adopted. Two types of LDPC codes, namely quasi-cyclic LDPC and spatially coupled LDPC, are considered. Simulation results show that, the proposed algorithm is able to accurately identify both types of LDPC codes no matterwhether an extra convolution code exists or not.