实时移动视频通信系统中的率失真优化编码传输

With the new study of 5G mobile communication standard and the next generation of video coding standard, mobile video communication with low delay and high quality is becoming a research hotspot. However, there are few theoretical researches for rate-distortion optimized coding and transmission with delay constraint in the mobile environment. This project focuses on the optimization of video coding and transmission under the environment with time-varying bandwidth and low-delay constraint. First, based on the R-λ model for rate control adopted in the up-to-date HEVC standard reference code, we propose an improved model with considering the effect of content characteristics; and propose to use mature machine learning methods instead of simple function modeling for regression analysis of relationships among model variables, in order to predict the adaptive λin the time-varying channels and select optimum combination of rate-distortion coding parameters under the rate constraint; Next, we modeling the relationship between channel coding redundant bit rate and λ, and with the respective mapping of source modeling and channel modeling, we implement a method to optimize the joint source channel coding in the λ domain; we further design the rate and delay control algorithms under time-varying transmission delay and bandwidth to achieve the global rate-distortion optimization while the resulted bit stream conforms to the low delay requirement suggested by standards. This research has practical value to the real-time high-quality mobile video communication applications.

随着5G移动通信标准和下一代视频编码标准开始研究制定，低延时高画质的移动视频通信成为一个研究热点。然而目前对于移动环境中延时受限的率失真优化编码传输的理论研究还比较缺少。本课题重点研究在时变带宽中低延时限制的视频编码和传输优化。首先基于最新的HEVC标准参考代码中码率控制所采用的R-λ模型，提出考虑了视频内容特性影响的改进模型；并提出用成熟的机器学习方法代替简单函数建模对模型变量关系进行回归分析，用于预测时变带宽中的自适应λ以及选择码率约束下的最优率失真编码参数组合；其次，建立信道编码冗余码码率在λ域的映射模型，并通过信源模型和信道模型各自的映射，实现在λ域的联合信源信道编码优化方法；进一步设计传输延时和带宽变化时的码率及延时控制算法，实现生成码流符合标准建议的低延时要求的全局率失真优化。本课题的研究对实时高质的移动视频通信应用有实际价值。

低时延高可靠是下一代移动通信的主要特点，然而移动通信系统中存在的多径衰落等特点造成物理信道的时变特性。由于受时变带宽、丢包、低延时限制等影响，对于实时高画质的移动视频通信中的视频编码传输优化设计目前仍然存在着诸多挑战。本课题的目的在于研究移动通信系统中低延时视频编码传输优化，对实时高画质的移动视频通信提出并实现一种可行的系统方案。首先基于最新的HEVC标准参考代码中码率控制所采用的R-λ模型，提出考虑了视频内容特性影响的改进模型;并提出用成熟的机器学习方法代替简单函数建模对模型变量关系进行回归分析，用于预测时变带宽中的自适应λ以及选择码率约束下的最优率失真编码参数组合。研究结果对HEVC编码器进行复杂度优化，有效权衡Rate-distortion (RD)性能损失与复杂度。完成基于内容自适应残差码率模型的HEVC帧级码率估计方法。研究了信道编码与传输方法研究，包括信道编码与网络缓存传输等。最终实现FPGA的视频编码传输验证平台。在该项目的支持下，一共发表了IEEE期刊论文9篇，IEEE会议论文6篇。本课题的研究通过信道自适应的视频编码设计提高移动通信系统资源的利用率，进而推动未来实时高画质移动视频通信应用，具有理论意义和实际价值。