Research On Joint Source-Channel Coding For Video Sources

Due to the design and implementation based on the source-channel separation theorem,the traditional framework of video coding and transmission cannot avoid the influence of "cliff effect",which is mainly in the form of the cliff-like decline of reconstructed video quality caused by errors and packet losses in wireless and network transmission,and thus makes the traditional scheme difficult to achieve the above goals.In order to reduce and eliminate the influence of the "cliff effect",the joint source-channel coding(JSCC)methods provide a variety of research directions and corresponding schemes.Nevertheless,these schemes still have different but urgent limitations that need to be solved.To overcome these limitations,two novel JSCC schemes have been proposed by this paper.Firstly,the related theories and methods of video coding and transmission have been introduced.The traditional joint source-channel decoding methods of arithmetic codes,and the joint source-channel parameter adjustment method on video coding and transmission,are mainly introduced.By introducing these aspects,the two novel joint source-channel coding methods proposed in the paper can be supported.Secondly,to reduce the great complexity raised by the traditional source-channel iterative decoding scheme,a novel joint source-channel decoding(JSCD)scheme designed with the idea of deep unfolding network structure,neural-based joint source-channel decoding(NJSCD),has been proposed.With an iterative JSCD framework for arithmetic codes(AC)designed by a customized neural network structure and a corresponding effective training method,the complexity of JSCD can be greatly reduced,and the performance of AC can be further improved.Furthermore,the bit-level decoding reliability for H.265/HEVC video decoding scenarios can be improved,and the influence of the "cliff effect"can be subsequently reduced by introducing NJSCD.Finally,to overcome the limitation of the existing joint source-channel parameter adjustment schemes trimming based on the channel condition,aiming at a cross-network transmission scenario,a novel real-time video source coding and transmission framework and the corresponding end-to-end joint source-channel parameter adjustment(E-JSCPA)mechanism have also been proposed.On the basis of the NJSCD scheme,the parameter adjustment mechanism can provide optimal video coding parameters,channel coding parameters,and neural network parameter selections of NJSCD decoder under the multiple constraints of both the wireless channel and the network transmission.It can effectively eliminate the influence of the "cliff effect",and ensure the reliability and low-latency performance for the real-time video encoding and transmission scenarios.The results show that the two JSCC schemes proposed in this paper can effectively reduce and eliminate the influence of the "cliff effect",At the same time,they can obtain better performance compared with other comparison schemes,and improve the practicability of the JSCC schemes used in the real video coding and transmission scenarios.