Caching and adaptive multiple description coding for fine-grained scalable video transmission

Abstract

Secondly, we study the interaction of fine-grained scalable video coding (SVC) and caching. Fine-grained scalable video is applied at intermediate caches to allow online video users to fetch video clips at different qualities. Also, a cache space allocation algorithm is provided to optimize the average PSNR performance on the users' side. Moreover, the work of scalable video caching is extended to two-cache scenarios. Besides the cache space allocation algorithm, exclusive-or (XOR) network coding is also introduced to combine the sending of data from the server to each cache to reduce the backhaul bandwidth consumption. Numerical results with actual YouTube and Netflix Prize data set input show that the algorithm and network coding not only provide improved luma PSNR performance, but also reduce the backhaul data traffic. Finally we extend the two-cache scalable video caching model to a more generalized multiple-cache model. The problem is solved by grouping caches into pairs, which simplify it to a two-cache network coding problem. Various cache pairing algorithms, including maximum weighted matching and the heuristic algorithm, are applied to optimize the backhaul traffic saving and numerical results show that the proposed pairing algorithms can achieve higher backhaul traffic saving than not having inter-cache cooperations.

The problem of fine-grained scalable video multicasting to heterogenous users is considered. In this problem the scalable video data is channel-coded through multiple description over forward error correction (MD-FEC) coding, sent from the server to the intermediate node and then split to multiple users. We designed an algorithm to perform the adaptive MD-FEC at the server side that would allow the intermediate node to send an appropriate part of the original data to users with different network environments. Numerical results show that the adaptive MD-FEC algorithm can provide higher average Peak Signal-to-Noise Ratio (PSNR) performance than layered multiple description coding (MDC), simulcast with FEC, and conventional point-to-point MD-FEC.