Channel-optimized vector quantization with mutual information as fidelity criterion

We consider the problem of channel-optimized vector quantization (COVQ) with mutual information as fidelity criterion. This problem is relevant in a communications context, where the goal is to maximize the end-to-end rate. We propose an algorithm which is similar to the information bottleneck method and solves the considered COVQ problem. In contrast to existing COVQ designs, the proposed method implicitly optimizes the labels of the quantizer outputs, thereby avoiding separate label optimization. The usefulness of the proposed method is corroborated by application examples and numerical results.