Quantization

Abstract

Various quantization techniques are available whose basic principles will be discussed in this chapter. For uniform quantization, the signal-to-noise ratio (SNR) is proportional to the signal level, hence it becomes smaller with decreasing signal power. However, especially in speech signals, small sample values are particularly frequent, corresponding to a probability density function (PDF), which can be approximated, for instance, by a Laplacian PDF, a gamma PDF, or by spherically invariant models. An alternative for reducing the dependency of the SNR on the (instantaneous) quantizer load is to use a uniform quantizer with K quantizer representation levels but with dynamical adaptation of the quantizer stepsize Δx. The chapter discusses scalar quantization. Vector quantization might be computationally very intensive, as the input vector x must be compared to all K code vectors x ^ i in order to minimize a distance measure. This case is called full search.

Authors 2

1. Peter Vary Aachen Institute of Communication Systems

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen University, Institute of Communication Systems, Aachen, Germany

2. Rainer Martin

   Ruhr University Bochum

   Affiliation as printed

   Ruhr-Universität Bochum, Institute of Communication Acoustics, Bochum, Germany