Motivation: In order to represent arbitrary signals by a finite number of bits, these signals are quantized using an appropriate quantization rule. Usually, these quantizers are designed with respect to the continuous signal and its distribution and the rate-distortion theory provides the required number of quantization bits in order to realize a pre-defined distortion measure between the analog signal and its digital representation. In contrast, with the information bottleneck method (IBM) quantizers are designed to maximize the mutual information between the quantizer output and a source signal of interest.
Goal: In this thesis the fundamental idea of the information bottleneck method is investigated. Based on a literature search the main theory is developed and appropriate quantization algorithms are implemented and tested. Depending on the progress, a limited number of applications will be selected to demonstrate the benefits of this novel method.
Requirements: In order to process this thesis, knowledge of Channel Coding I, Communication Technologies, and programming skills in Matlab are essential. |