ABSTRACT
We know that a bandlimited analog signal can be recovered from its sampled values as long as the sampling rate satisfies the Nyquist criterion. For transmitting the information, however, the infinite-precision sampled values have to be rounded off to a finite number of bits. This process, referred to as quantization, is the principal source of signal degradation in waveform coders. This degradation, referred to as quantization noise, depends on the quantizer characteristics as well as the input signal statistics. We will derive analytical expressions for quantifying this noise and the achievable SNR for both uniform and nonuniform quantizers. However, the main goal of this chapter is to explain the theoretical basis for the μ-law and A-law logarithmic quantizers that have been standardized for speech coding. Finally, we touch upon the topics of optimal and adaptive quantizers, which are actually used in the differential coding standards discussed in the next chapter.
