Streaming Codes with Unequal Error Protection against Burst Losses
Streaming Codes with Unequal Error Protection against Burst Losses
复制标题
具有不等错误保护的流代码，防止突发丢失
DOI：
--
复制
发表时间：
2018
期刊：
影响因子：
--
通讯作者：
J. Apostolopoulos
中科院分区：
文献类型：
--
作者：
Mahdi Haghifam;Ahmed Badr;A. Khisti;Xiaoqing Zhu;Wai;J. Apostolopoulos
用于实时交互式应用（如音频和视频流）的差错控制码必须在严格的延迟约束下操作，并且对突发丢失具有弹性。以前的作品，其特征在于最佳的代码，保证完美的恢复时，突发丢失的所有源数据包是低于一定的最大阈值。在这项工作中，我们介绍了两个概念的不等错误保护（UEP）的流代码。第一种是符号级UEP，其中每个源分组由两个子分组组成，其中一个子分组具有比另一个更高的重要性级别。虽然重要的子分组具有相同的截止日期，但它必须从更长的突发丢失中恢复。如果突发丢失低于某个标称阈值，则需要完整恢复整个源分组。第二个概念是包级UEP。在这里，我们假设在奇数和偶数时间到达的源数据包具有不同的重要性级别。当突发长度低于某个阈值时，必须恢复所有源分组。另一方面，在较长的突发期间，我们只需要在偶数时隙恢复源分组。我们讨论了实际的动机，这两个设置和开发的编码方案，使用以前提出的流代码作为组成代码。我们建立最优性或近最优性保证，通过信息论匡威。在吉尔伯特信道上的仿真表明，这些码在很宽的信道参数范围内优于基线方案。
Error control codes for real-time interactive applications such as audio and video streaming must operate under strict delay constraints and be resilient to burst losses. Previous works have characterized optimal codes that guarantee perfect recovery of all source packets when the burst loss is below a certain maximum threshold. In this work we introduce two notions of unequal error protection (UEP) in streaming codes. The first is symbol-level UEP where each source packet consists of two sub-packets, one of which has higher level of importance than the other. While the important sub-packet has the same deadline it must recover from a longer burst loss. Perfect recovery of the entire source packet is required if the burst loss is below a certain nominal threshold. The second notion is packetlevel UEP. Here we assume that the source packets arriving at odd and even times have different levels of importance. When the burst length is below a certain threshold, all the source packets must be recovered. On the other hand, in the period of the longer burst, we only require the recovery of the source packets at even time slots. We discuss practical motivations for both settings and develop coding schemes that use previously proposed streaming codes as constituent codes. We establish optimality or near optimality guarantees through information theoretic converse. Simulation over Gilbert channels show that these codes outperform baseline schemes over a wide range of channel parameters.