Distributed coding of spatio-temporally correlated source

This dissertation studies certain problems in distributed coding of correlated sources. The first problem considers the design of efficient coders in a robust distributed source coding scenario. Here, the information is encoded at independent terminals and transmitted across separate channels, any of which may fail. This scenario subsumes a wide range of source and source-channel coding/quantization problems, including multiple descriptions and the CEO problem. A global optimization algorithm based on deterministic annealing is proposed for the joint design of all the system components. The proposed approach avoids many poor local optima, is independent of initialization, and does not make any simplifying assumption on the underlying source distribution.;The second problem considered is of scalable distributed source coding. This is the general setting typically encountered in sensor networks. The conditions of channels between the sensors and the fusion center may be time-varying and it is often desirable to guarantee a base layer of coarse information during channel fades. This problem poses new challenges. Multi-stage distributed coding, a special case of scalable distributed coding, is considered first. The fundamental conflicts between the objectives of multi-stage coding and distributed quantization are identified and an appropriate design strategy is devised to explicitly control the tradeoffs. The unconstrained scalable distributed coding problem is considered next. Although standard greedy coder design algorithms can be generalized to scalable distributed coding, the resulting algorithms depend heavily on initialization. An efficient initialization scheme is devised which employs a properly designed multi-stage distributed coder. The proposed design techniques for multi-stage and unconstrained scalable distributed coding scenarios offer substantial gains over naive approaches for multi-stage distributed coding and randomly initialized scalable distributed coding respectively.;The third problem considered is distributed coding of sources with memory. This problem poses a number of considerable challenges that threaten the practical application of distributed coding. Most common sources exhibit temporal correlations that are as important as inter-source correlations. Motivated by practical limitations on both complexity and delay, especially for dense sensor networks, the problem is re-formulated in its fundamental setting of distributed predictive coding. The most basic tradeoff (and difficulty) is due to the conflicts that arise between distributed coding and prediction, wherein 'standard' distributed quantization of the prediction errors, if coupled with imposition of zero decoder drift, drastically compromises the predictor performance and hence the ability to exploit temporal correlations. Another challenge arises from instabilities in the design of closed loop predictors in distributed coding setting. These fundamental tradeoffs in distributed predictive coding are identified and a more general paradigm, is proposed where decoder drift is allowed but explicitly controlled. The proposed paradigm avoids the pitfalls of naive techniques and produces an optimized low complexity and low delay coding system.