| Design of robust system deals with reducing the variance of system response. In this thesis, using the four parameters, signal, control, noise parameter, and response from Robust Design, a procedure for systematic robust system design is applied to video coder optimization.; Design of robust system is implemented with the proposed two models. One of them uses Volterra series that can be effectively used for designing nonlinear system. This model also shows all the effects of combinations of parameters, enabling us to analyze the problem of reducing number of experiments. The other model is developed using a LTI system with state-space representation. Using the Kalman decomposition, a LTI system can be decomposed by a controllable and an uncontrollable system. The uncontrollable system can be effectively modeled as the uncontrollable source of variation. Furthermore, from the effect of parameters, the model is expanded to analyze the interaction of the parameters.; In robust system design with these two models, we show when the number of experiments can be reduced by design of experiment technique. If we know whether there is any interaction between parameters, we can reduce the number of experiments by a few orders of magnitude in some cases.; The H.263 video coder is used for the proposed method. Most video coding standards provide only a generalized syntax and semantics framework for video coders, leaving the selection and optimization of the right parameter set to the responsibility of implementators. The choice of the right parameter set is, however, quite difficult. This difficulty is particularly amplified in the low bit rate video coding. The robust parameter set is obtained using the proposed method. A number of experimental results, using the obtained parameter set, confirm the improvement over a wide variety of input data streams. |
