Signal coding, as the process of obtaining a suitable digital representation of signals, is an important aspect of modern telecommunication environment. To attain desirable signal quality, keeping the relevant signal attributes, while diminishing the signal distortion, different quantizers, their parametrization and coding methods, are suggested. In this paper we address the iterative method for the support region threshold determination, as the segment of good parametrization for robust μ-law quasilogarithmic quantizer, designed for the Laplacian source of unit variance. The simple solution that we propose, overcomes the shortcomings and impediments that can be influenced by the unpredictable nature of speech and audio signals, approximated by the Laplacian probability density function. With the increasing widespread quantization deployment in many contemporary applications, the effectiveness of the proposed robust quasilogarithmic quantizer solution can be recognized as the part of the ubiquitous optimization task. A growing interest in deep neural network (DNN) is directed towards to the efficient inferencing using the compressed models.

Keywords: Support region, Clipping, Quasilogarithmic quantizer parametrization, Curve overlapping