Robust predictive quantization: a new analysis and optimization framework

This work is focused on computing-via a deterministic optimization with linear matrix inequality (LMI) constraints, rather than a pseudorandom simulation-the performance of predictive quantization schemes under various scenarios for loss and degradation of encoded prediction error samples. The ability to make this computation then allows for the optimization of prediction filters with the aim of minimizing overall mean squared error (including the effects of losses) rather than to minimize the variance of the unquantized prediction error sequence. The main tools are recent characterizations of asymptotic state estimation error covariance and output estimation error variance in terms of LMIs. These characterizations apply to discrete-time jump linear systems in which the discrete portion of the system state is a Markov chain. Translating to the signal processing terminology, this means that the signal model is "piecewise ARMA," as is standard in many forms of speech processing.