Joint source-channel vector quantization using deterministic annealing

We present a new approach to the combined sourcechannel vector quantizer design problem. The method utilizes deterministic annealing to avoid local minima that trap conventional descent algorithms. The temperature here is used to control the fuzziness of the encoder and decoder association probabilities. In the low temperature limit, our method reduces to a descent method that is an analogue of the Generalized Lloyd Algorithm (GLA) for noisy channels. Thus, in this sense, it is a generalization of noisy channel GLA (NC-GLA). We performed simulations to compare our approach with several versions of NC-GLA. Our method outperformed the other methods under all test conditions. Moreover, the gains over other methods grow with the codebook size.