Multiple description coding and path diversity for video communication over packet loss networks

This thesis report describes a novel multiple description coding (MDC) scheme that falls under the joint source channel coding paradigm. We have designed and implemented (using Matlab and C++) the core framework for multiple description coding to transmit images over the packet loss networks. Useful contribution to this work includes the rate allocation algorithm based on the packet loss estimates of the channel. Due to inherent scalability support by the source encoder we have extended our framework for delivering video. The MDC scheme results better average than the SDC by 2dB at 10% packet loss and importantly the variance of MDC scheme is negligibly smaller compared to SDC scheme at similar conditions. This enables to produce smoother playback of the video in the MDC case. This system provides good quality in the descriptions themselves and also provides graceful degradation of reconstructed image quality with the loss of descriptions. Key feature of our proposed approach is that it gives fine granular control over the description formation enabling dynamic scalable MDC streams based on the channel conditions. In another experiment we studied the performance of rate allocation strategy for both random bit and packet loss error channels in conjunction with the JPEG2000 error resilience features. The performance improvement of such a design does not offer significant benefits suggesting that MDC is more valuable in packet loss scenarios compared to bit error channels. We believe that diverse path selection needs to be integrated to make this proposed MDC scheme more beneficial to image and video transmission. We have integrated multiple description coding framework with path diversity algorithm (Eppstein algorithm) to study the performance of video transmission over multi-hop packet loss networks. Experimental results show our MDC design provides higher performance gain in comparison with the other approaches proposed in the literature when used with suitable rate allocation conditions. We also demonstrate that our heuristic scheduling algorithm (carrying layered compressed video stream payload) outperforms Greedy algorithm and the conventional EDF algorithm in trace-driven simulations. This higher quality was achieved at the cost of increased time-complexity. (Abstract shortened by UMI.)...