A Dual-Mode 2-bit Uniform Scalar Quantizer for Data with Laplacian Distribution
- Resource Type
- Conference
- Authors
- Peric, Zoran; Denic, Bojan; Jovanovic, Aleksandra; Savic, Milan; Vucic, Nikola; Nikolic, Anastasija
- Source
- 2021 29th Telecommunications Forum (TELFOR) Telecommunications Forum (TELFOR), 2021 29th. :1-4 Nov, 2021
- Subject
- Communication, Networking and Broadcast Technologies
Components, Circuits, Devices and Systems
Computing and Processing
Engineering Profession
Fields, Waves and Electromagnetics
General Topics for Engineers
Photonics and Electrooptics
Signal Processing and Analysis
Transportation
Quantization (signal)
Laplace equations
Image coding
Neural network compression
Dynamic range
Robustness
Telecommunications
image classification
neural network
quantization
source coding
- Language
In this paper, a novel dual-mode 2-bit scalar quantizer based on uniform quantization of data with Laplacian distribution is proposed. It is composed of two uniform quantizers with different step size, whereby the rule for quantizer selection is determined in order to provide the highest signal to quantization noise ratio in the given range of data variance. The proposed dual-mode scalar quantizer also performs frame-by-frame processing. It is shown that in terms of signal to quantization noise ratio our proposal is able to outperform single uniform quantizer in a wide dynamic range of input data variances. To verify benefits achieved with the proposed dual-mode quantizer the experiment is performed with the Multi-Layer Perceptron (MLP) weights.