Empirical quantizer design in the presence of source noise or channel noise

The problem of vector quantizer empirical design from training vectors is studied for noisy channels and for noisy sources. It is shown that global empirical error minimization for designing quantizers for transmission over a discrete noisy channel have the same performance (convergence rate) as in ordinary quantizer design. For noisy source quantization an appropriate analogue of empirical error minimization is developed. Consistency and convergence rates are proved under appropriate regularity conditions in this case