Robust vector quantizer design by noisy channel relaxation

This article proposes a method to design a vector quantizer (VQ) for robust performance under noisy channel conditions. By re-optimizing the quantizer at progressively lower levels of assumed channel noise, the design is less susceptible to poor local optima. The method is applied to: (1) channel-optimized VQ design; and (2) index assignment for a source-optimized VQ. For both problems, we demonstrate substantial performance improvements over commonly used techniques