Research On Channel Code Identification Based On Deep Learning

Channel code identification refers to the task of identifying the channel code used by the received signal and has a wide range of applications in non-cooperative communications and intelligent communications.Most of the existing channel code identification algorithms are based on traditional signal processing methods,and suffer from the problems of high complexity and poor accuracy.In recent years,deep learning has been developing rapidly and is known as one of the most breakthrough techniques in the field of artificial intelligence.This thesis focuses on the implementation of deep learning in channel code identification,which makes use of the powerful classification ability of deep learning to improve the identification performance.First,this thesis discusses the existing channel coding techniques and recognition methods,and introduces the well-known deep neural networks and deep learning frameworks.On this basis,Low-Density Parity-Check(LDPC)code and convolutional code are taken as the examples,and convolutional neural network as well as Tensor Flow framework is considered for later study.Secondly,this thesis studies the deep learning based channel code identification with hard decision,in which the signal to be identified is a bitstream after demodulation.To solve this issue,the thesis exploits word vector representation and proposes a channel coding recognition algorithm based on Text CNN.This algorithm treats the received bitstream as a text sequence,represents the sequence in the form of a word vector,converts the vector into a sentence matrix,and feeds the matrix into Text CNN.Experimental results show that the proposed algorithm is able to effectively distinguish between Quasi-Cyclic(QC)and Spatially-Coupled(SC)LDPC codes,and accurately identifies LDPC codes and convolutional code at high signal-to-noise ratios.Finally,this thesis studies the deep learning based channel code identification with soft decision,where the signal to be identified is a soft information sequence.To conquer this problem,the thesis adopts time-domain sequence representation and designs a channel code identification algorithm based on Fully Convolutional Network(FCN).The algorithm regards the channel code identification task as a classification problem in time domain,and uses FCN to directly learn from the features of the soft information sequence and consequently infer the channel code.Experimental results show that the proposed algorithm can achieve the identification of LDPC codes as well as convolutional code and has relatively low model complexity.