Algorithms for optimal multiple description codes design

现代通信技术的出现对多媒体传输提出了新的挑战。在有线和无线网络中都遇到的挑战之一是必须减轻数据包丢失的影响。多描述编码（multiple description coding， MDC）是有效解决这一问题的一种范式。MDC的基本思想是创建信号的几种描述，这样可以从每个单独的描述中重构出具有一定保真度的信号源，当有更多的描述可用时，它们可以相互改进，从而实现更高保真度的重构。提出的研究的主要目标是开发有效的算法来设计最优的多重描述码。重点将放在MDC的两种方法上：多描述标量量化和可扩展码流的不均匀擦除保护的分组。除了为先前制定的优化问题开发新的解决方案（更快和/或更高的精度）外，申请人还将研究新的MDC框架，并将为从实际关注中提出的新问题提出解决方案。提出的研究将导致在有损网络上实现鲁棒多媒体传输的新技术，从而更有效地利用传输速率上的可用资源。这些技术将在实践中直接适用。此外，本研究将为MDC设计领域带来新的见解，从而有助于提高该领域的知识和理解，并为进一步的研究奠定基础。

The advent of modern communications raises new challenges for multimedia transmission. One of these challenges, encountered in both wired and wireless networks, is the necessity to alleviate the impact of packet loss. A paradigm to effectively solve this problem is multiple description coding (MDC). The basic idea in MDC is to create several descriptions of the signal such that the source can be reconstructed to a certain fidelity from each individual description, and when more descriptions are available, they can refine each other, leading to a higher fidelity reconstruction.The main objective of the proposed research is the development of efficient algorithms for the design of optimal multiple description codes. The emphasis will be on two approaches for MDC: multiple description scalar quantization and uneven erasure protected packetization of scalable codestreams. Beside developing newsolutions (faster and/or of higher accuracy) to previously formulated optimization problems, the applicant will investigate new MDC frameworks and will propose solutions to new problems raised from practical concerns.The proposed research will lead to new techniques for robust multimedia transmission over lossy networks, which more efficiently use the available resources in transmission rate. These techniques will be directly applicable in practice. Additionally, this research will bring new insights in the area of MDC design, thus contributing to the advancement of knowledge and understanding in the area and laying background for further research.