| Causal source coding is widely used for low-latency compression of real-world audio and video signals. The most prominent class of causal source codes used in this context are codes based on differential predictive coding. Differential predictive coding is ill-suited, however, for coding scenarios characterized by encoder uncertainty regarding decoder reconstructions. In this thesis we present a coset code based causal coding framework, motivated by the Wyner-Ziv coding methodology for the information theoretic decoder side-information paradigm, for source compression in such scenarios. In particular, this thesis addresses the problems of two-channel predictive multiple description coding, and compression for streaming of digital walkthrough data. We demonstrate how coset code based constructions can be used to provide efficient, low-latency compression for these problems. Evaluation of the presented constructions, on idealized and real-world video sources, confirms the promise of the proposed framework. |
