面向软传输的视频编码技术研究

Cliff effect is one big issue in wireless visual communication system. In this proposal, we focus on this issue and propose to study a new kind of video coding method for softcast. Different from traditional method, the video signal is not coded into sensitive binary streams but robust real-valued sequences in video softcast. These robust real-valued sequences can be directly transmitted in the PHY layer of wireless channel without any further FEC protection. More importantly, softcast is lossless coding and its reconstruction distortion fully depends on the channel noise. This means softcast has no cliff effects, and therefore avoids a lot of problems in traditional system, including the source channel mismatch problem due to unstable channel noise, the rate decision problem due to unknown channel noise, and the difficulty of accommodating multiple users in broadcasting. Since softcast is very new, there are a lot of problem to solve, such as the transform coding problem, the predictive coding problem, the power allocation and control problem, the power-distortion optimization problem and etc. The solution of these problem will promote the application of softcast in wireless communication, as well as the advance of corresponding theory and methods. Therefore, this direction is of great importance and is worth to study.

本研究主要针对无线视频编码传输中的悬崖效应问题，研究面向软播的视频编码方法。与传统方法不同，无线视频软播系统中视频不是被编码为敏感的二进制码流，而是有高度抗噪能力的实数序列。更重要的是视频软播属于无损编码，且最终的失真取决于信道噪声，因而可以有效杜绝悬崖效应，从而在根本上解决由信道质量不稳定引起的信源信道不匹配，由信道质量未知引起的码率控制困难，以及广播信道中各用户难以兼顾等问题。由于目前国内外对于视频软播的研究均处于起步阶段，其中很多问题仍急需解决，比如视频信号的变换表示、预测编码、能量分配与控制以及能量失真优化技术等。这些问题的解决，将使视频软播在无线视频传输的应用从潜在变为现实，并且会促进各种对于新应用的探索，以及促进视频编码理论和方法的发展。因此具有重要的理论意义和广泛的应用价值，值得深入展开。

数字通信一般来说具有抗干扰能力强的优点，但是一旦出错会影响视频信号解码，导致马赛克或者无法解码。模拟通信则正好相反，抗干扰能力弱，但是没有悬崖效应，可充分利用信道。近年来的研究发现，用数字方式中高密度调制发射模拟信号得到的数模混合传输（即软传输），可以同时具有数字和模拟的优点。我们研究了一系列面向软传输的视频编码技术，包括高效的面向软播的视频变换表示，高效的面向软播的视频预测编码，高效的面向软播的传输能量分配方法。提出了基于压缩感知的软播、高光谱卫星图像软播等视频软传输、分层软播等基于软传输的视频编码框架，以及变形块运动信息获取的merge技术、复合局部非局部帧内预测技术等一系列视频编码技术。 共发表SCI期刊论文13篇，EI会议论文14篇，包括IEEE transactions论文5篇，CCF A类会议论文1篇，B类会议论文2篇。发明专利5项。6项技术提案被AVS标准接收，6项技术被FVC标准接收。