Research On The Key Technologies For Joint Source-Channel Coding

With the continuous development of mobile communication technology,users’ demands for communication quality are also increasing.Due to transmission noise in the channel and imperfections in the signal source,transmission errors can occur due to factors such as transmission noise and channel fading,which can affect communication quality.The source-channel coding technology can improve communication quality and transmission efficiency.Therefore,the study of source-channel coding technology has important practical significance and theoretical value.Firstly,this dissertation introduces the related principles of sourcechannel coding,including sparse vector coding method based on lowdensity parity-check(LDPC)code and source feature extraction method based on deep learning.These principles provide support for the construction of Double Protograph Low-density Parity-check(DP-LDPC)code based on 5G New Radio(NR)and image-video source joint coding method based on Swin Transformer proposed in later sections.Secondly,this dissertation proposes a source-channel joint coding method based on DP-LDPC.DP-LDPC can achieve high code rate and low bit error rate with low complexity and short decoding time.However,traditional DP-LDPC codes have the drawback of difficult construction and inability to adapt to various source entropy rates.In order to overcome these difficulties,this dissertation designs a construction method based on 5G NR.In this method,the entropy rate of the source is first evaluated,and the appropriate source coding rate is selected through table lookup.Then,the source coding base matrix is cropped from the LDPC code base matrix of 5G NR using a specific method.Finally,the source coding base matrix is combined with the channel check base matrix to construct the joint check base matrix and the final sourcechannel joint coding system.Experimental results show that the DPLDPC coding method can achieve a performance gain of about 1dB compared to the traditional separate source-channel coding.In addition,this dissertation also designs a deep learning sourcechannel joint coding method based on Swin Transformer.Swin Transformer is an efficient deep learning model that can effectively improve the speed and accuracy of image and video processing.This dissertation applies Swin Transformer to source-channel joint coding to realize a mixed digital-analog transmission scheme for video and image sources.Meanwhile,this dissertation estimates the entropy rate of the image using the information bottleneck and uses the hierarchical window feature of Swin Transformer to achieve hierarchical extraction of source features and adaptive rate coding.Experimental results show that this method can save about 70%of channel resources and has better antinoise performance under adverse channel conditions.In conclusion,this dissertation proposes a source-channel joint coding scheme based on DP-LDPC codes and Swin Transformer.These technologies not only have good theoretical performance,but also have broad application prospects.