Research On Joint Source Channel Coding Based On Machine Learning

In the multimedia coding and transmission system, it is contradictory between source coding to obtain coding efficiency though compressing and channel coding to increase reliability by adding redundancy.Sometimes, the optimal combination of them is not always able to obtain the global optimum of the system. If influence of channel environment,usage scenarios for different channel environment as well as the advantages of JSCC could be fully considered and used, the algorithm will achieve a higher performance gain.This paper focuses on the transmission of compressed video streams in wireless channel combined with the advantage of clustering algorithm in classifying the channel environment and the bit stream structure of HEVC.Firstly, this paper deeply studies a variety of application strategies of joint source channel coding (JSCC) and uses k-means++ clustering algorithm to cluster the different channel conditions based on the analyses of various factors which have influence on wireless channel. The proposed scheme can achieve optimal bit allocation scheme adaptively by using clustering results combined with minimum end-to-end distortion principle, based on channel forward feedback methods in traditional JSCC scheme. Also the proposed scheme can reduce the computational complexity of end-to-end distortion.Secondly, this paper selects the High Efficiency Video Coding(HEVC) as source coding. According to the characteristics of HEVC time domain classification, more bit will be allocated to the important part of the flow, such as public information which includes syntax element of video parameter set, sequence parameter set and picture parameter set, as well as lower time-domain layers. The scheme of unequal error protection is used to further enhance system performance.Finally, experiment results show that, the proposed algorithm could achieve a better performance by adjusting the bits allocated to the source and channel dynamically. In addition, a higher gain will be obtained combined with unequal error protection.