Research On Encoding And Decoding Methods Of Protograph-based LDPC Codes

In mobile communication systems,the effectiveness and reliability of transmission has always been the pursuit of system design.Channel coding techniques are essential to ensure reliability.Among many channel coding techniques,low-density parity-check(LDPC)codes have become a mainstream channel coding technology by virtue of effective encoding and excellent decoding performance,and been applied to various practical systems,such as Wi-Fi,Wi-MAX,fifth-generation mobile communication(5G)systems.However,there is still room for improvement in the encoding and decoding method of existing LDPC codes.Therefore,it is necessary to study it to further improve system reliability.In modern LDPC codes,the parity-check matrix generally adopts a quasi-cyclic(QC)structure.The code with this structure is called a QC-LDPC code.Most of the current QC-LDPC codes with excellent performance are constructed from protographs.Therefore,this type of QCLDPC codes constructed by the protograph is also called a protograph-based LDPC(PB-LDPC)code.In order to improve the coding performance of existing PB-LDPC codes,this thesis studies the design and optimization from three aspects: construction methods & encoding structures,protographs,and decoding algorithms.The main research work and contributions include the following:Firstly,based on the given protographs,we improve the expanded codeword performance from two aspects: construction methods and encoding parity structures.In terms of the construction methods,for the problem that weights of the protograph nodes of QC-LDPC codes constructed by the single-weight circulants(SQC-LDPC codes)are limited at high code rates,we propose a construction method of QC-LDPC codes with multi-weight circulants(MQCLDPC codes).In order to support multi-weight circulants construction,this thesis introduces a method to prevent multi-weight circulants from generating short cycles in the error minimization progressive edge growth algorithm.When the expanding factor is fixed,the MQC-LDPC code can obtain a better protograph structure,codeword weight distribution,and stopping set weight distribution than the conventional SQC-LDPC code.The experimental results show that,in the additive white Gaussian noise channel,the constructed MQC-LDPC codes have better performance compared to the high rate SQC-LDPC codes in the Wi-Fi standard and the Wi-MAX standard whether in the high or low signal-to-noise ratio(SNR)region.In terms of encoding parity structures,for the problem that low-degree nodes cause poor error floor performance,we propose a block triple diagonal parity structure with all nodes of degree 3 for QC-LDPC codes.This block triple diagonal parity structure without low degree nodes can effectively improve the error floor performance and support low complexity encoding.Using this structure and modulo-lifting method,we constructed a series of QC-LDPC codes with scalable code lengths.The simulation results show that our codewords can achieve better performance in the error floor region than the existing scalable codewords.Then,for the design problem of the low-rate LDPC code with short code lengths and fast convergence in the ultra-reliable low-latency communication application of 5G systems,we propose a series of rate compatible protographs with raptor-like structures.These protographs were derived from the low rate codes given by Richardson-Urbanke(RU).In the proposed protograph,we use the protograph-based extrinsic information transfer(PEXIT)analysis method with limited iteration numbers to gradually optimize the connection degrees.By the optimized protograph,we constructed rate compatible LDPC codes and simulated them.The simulation results show that the proposed LDPC codes have significant performance advantages compared with the original RU low rate codes,the traditional AR4 A low rate codes,and the low rate codes given by Qualcomm in the 5G system standardization proposal when the code length is short and the number of decoding iterations is small.Finally,for the performance degradation problem of the traditional normalized min-sum(NMS)and offset min-sum(OMS)decoding algorithms at low code rates,we propose a generalized linear-approximation min-sum(LAMS)decoding algorithm for the PB-LDPC codes and use a neural network to optimize the factors.The proposed LAMS algorithm linearly adjusts the magnitudes of the check node update and channel input by introducing multiplicative and additive factors for each iteration.These factors are optimized by a small shallow neural network using training data generated by the LAMS decoder in each iteration.We optimized the LAMS decoder for the PB-LDPC code in the 5G system and gave the linear factors for each iteration.The simulation results show that the LAMS algorithm shows a significant performance improvement compared with the traditional NMS and OMS algorithms,and even better performance than the belief propagation(BP)algorithm in some high SNR regions.