Implementation Of Non-Binary LDPC Codes And Research On Modulation Pattern Recognition

Channel coding and modulation are two important parts of digital communication system.Channel coding improves the reliability by adding redundancy to realize error detection and correction.By using different modulation modes,the compromise of power efficiency,frequency band efficiency and reliability can be achieved.The error correction performance of LDPC codes can approach Shannon limit.Non-binary(NB)LDPC codes have better burst error resilience and higher transmission.In addition,in the communication system,in order to correctly demodulate and recover the source information,the receiver must predict the modulation mode of the transmitter.In non-cooperative communication or intelligent communication system,the receiver needs to identify the modulation mode according to the signal received by the channel.With the continuous development of in-depth learning,signal modulation pattern recognition based on in-depth learning method has attracted wide attention because of its high efficiency and recognition accuracy.Based on the construction of the check matrix of the non-binary quasi-cyclic(QC)LDPC codes,this thesis investigates the realization of the encoding and decoding algorithm of the non-binary LDPC codes.At the same time,a deep convolution neural network model is established based on deep learning.The modulation pattern recognition algorithm is studied and simulated.The main works are as follows:(1)On the basis of finite field,a check matrix of non-binary QC-LDPC codes is constructed,and the encoding and decoding algorithms are thoroughly investigated.Based on the basic matrix of IEEE802.16e standard,a binary QC check matrix is obtained according to the code lenegth and code rate.Then,a finite field with the same order as the binary QC-LDPC code is constructed.Furthermore.it could obtain the non-binary quasi-cyclic check matrix by randomly replacing the non-zero elements in the check matrix.Because the check matrix has the characteristics of quasi-cyclic and quasi-double diagonal lines,it can be used to encode the information sequence directly.Because of avoiding the complex calculation in the process of transforming the check matrix to the generation matrix.it is more advantageous to realize fast encoding.(2)Aiming at the coded modulation modes of non-binary LDPC codes,the conversion rules between bits and symbols are studied when the order of codes is different from that of the modulation.while the corresponding calculation methods of decoding initialization message are given in this thesis.Moreover.belief propagation decoding algorithm based on fast Hadamard transformation(FHT)is thoroughly studied.and the expansion methods of Hadamard matrices with different orders are presented.The complexity of the updating of the check nodes is reduced with the use of FHT.and then the decoding speed is improved.At the same time,in the proceeding of-decoder implementation.the software storage structure is optimized by only stored the values of non-zero elements in the matrix and their location information.which can reduce the memory overhead.(3)Based on deep learning,a deep convolution neural network model is constructed to study the modulation pattern recognition.The neural network model is composed of 12 layers,and the specific parameters of each layer are given.Furthermore.the black-and-white constellation is colored by using the aggregation degree of the received signals in different regions of the constellation.The experimental results show that the colored constellation image can obtain the higher recognition accuracy owing to the more distinct distinguishing features.When SNR is 5dB.the recognition rates of BPSK.QPSK,16QAM,32QAM and 64QAM are all above 92%.