Channel optimized predictive VQ
J. Linden · 1999
Combined source-channel coding is considered for the case of predictive vector quantization. A design algorithm for channel optimized predictive vector quantizers is proposed. Under reasonable assumptions, the optimal encoder is presented and a sample iterative design method that simultaneously optimizes the predictor and the codebook is derived. We also demonstrate that this design method can be used to obtain index assignments that are advantageous to what is obtained by post process index assignment algorithms. Results are presented for a correlated Gauss-Markov process and for speech LSF parameters.