The Design Of LDPC Decoder Based On Deep Learning

In order to explore the application of deep learning in LDPC(Density Parity Check Low),this paper puts forward the LDPC decoding algorithm based on deep learning,and analyzes its feasibility.As a kind of mature deep learning model,deep belief networks could achieve better performance on data classification.LDPC codes is linear block codes.The decoding process is similar to the process of data classification.LDPC decoding process can make use of the classification ability of DBN.By training the DBN model,the model can be used to decode LDPC codes.Experimental results show that the LDPC decoding algorithm based on deep learning can improve the performance of short code in LDPC.At the same time,the decoding model used in BCH code and Hamming code,have a certain performance promotion.Compared with the traditional neural network decoding,the decoding algorithm based on deep learning has the following advantages:1)deep learning decoding is a universal model,and for different codes,we only need to adjust the network size and depth;2)training speed is fast,and deep learning by greedy layer pretraining can make the model parameters close to the global optimal decoding model,avoiding the local optimum;3)the network capacity of traditional neural network is small,and the capacity model of deep learning network is very large;4)correcting ability of deep learning model has better performance.Based on the idea of combination in LDPC decoding algorithm and deep learning technology,this paper studied a new method to decode LDCP codes based on deep learning.The work in this paper mainly focuses on the following aspects:(1)Analysis of decoding algorithm of LDPC codes by using deep learning as a classifier.Designed a LDPC decoder based on deep learning.The feasibility of LDPC decoding algorithm based on deep learning is proved,and when SNR is very small,this algorithm could achieve better performance than the traditional LDPC decoding algorithm.(2)The effects of different DBN model parameters,such as the learning rate,the hidden layer size,etc.,are analyzed.(3)Analysis of decoding algorithm of block code by using deep learning as a classifier.BCH code and Hamming code are studied in this paper,and the decoding algorithm of BCH code and Hamming code based on deep learning is implemented.Experiments showthat,in the low SNR,decoding performance based on deep learning is better than the hard decision decoding performance of hamming code and the hard decision decoding performance of BCH codes.