Communication systems form a defining aspect of modern society. Particularly important in this respect is the transmission of audio-visual information, since it is a fundamental means of human interaction. The rapid growth in network capacity and the increasing diversity of services is associated with a rapidly increasing heterogeneity of network attributes and quality of services (QoS). Current state-of-the-art coding systems for audio and video are designed for a particular scenario, and will fail if faced with a communication channel with different properties.
In contrast, we propose to introduce flexibility into source and channel coding algorithms. To this purpose we will develop a new paradigm that replaces off-line coder design by real-time coder design, thus ensuring the optimality of the coder over a large range of environmental conditions (e.g., rate, quality, and robustness requirements imposed by the channel). The new algorithms are expected to retain or improve on the efficiency of current state-of-the-art algorithms.
The new paradigm is based on the usage of universal statistical models of the sources and channels, and generic models of distortion. The flexibility is obtained through the use of models that allow analytically derived source and channel coding methods that do not require training.
Funding SchemeSTREP - Specific Targeted Research Project