A Graph-Based Message Passing Approach for Joint Source-Channel Coding via Information Bottleneck Principle

In this paper, we focus on an extended version of noisy source coding wherein the compressed data shall be transmitted over an imperfect (forward) channel for further processing. As the quantization design framework, we deploy the Information Bottleneck principle and propose a novel treatment by successful exploitation of a quite generic and highly flexible graph-based clustering routine known as Affinity Propagation. We also provide simulation results regarding a typical digital transmission setup to compare the performance of our proposed treatment with a state-of-the-art routine from literature.