Audio Signal Processing and Coding

3.1 INTRODUCTION

This chapter provides an introduction to waveform quantization, and entropy coding algorithms. Waveform quantization deals with the digital or, more specifically, binary representation of signals. All the audio encoding algorithms typically include a quantization module. Theoretical aspects of waveform quantization methods were established about fifty years ago [Shan48]. Waveform quantization can be: i) memoryless or with memory, depending upon whether the encoding rules rely on past samples; and ii) uniform or nonuniform based on the step-size or the quantization (discretization) levels employed. Pulse code modulation (PCM) [Oliv48] [Jaya76] [Jaya84] [Span94] is a memoryless method for discrete-time, discrete-amplitude quantization of analog waveforms. On the other hand, Differential PCM (DPCM), delta modulation (DM), and adaptive DPCM (ADPCM) have memory.

Waveform quantization can also be classified as scalar or vector. In scalar quantization, each sample is quantized individually, as opposed to vector quantization, where a block of samples is quantized jointly. Scalar quantization [Jaya84] methods include PCM, DPCM, DM, and their adaptive versions. Several vector quantization (VQ) schemes have been proposed, including the VQ [Lind80], the split-VQ [Pali91] [Pali93], and the conjugate structure-VQ [Kata93] [Kata96]. Quantization can be parametric or nonparametric. In nonparametric quantization, the actual signal is quantized. Parametric representations are generally ...