Task-Oriented Single- and Multi-Source Semantic Source Coding

Semantic communication aims at conveying semantic features rather than transmitting lossless data. The development of semantic communication is greatly hindered due to the lack of a general mathematical model for semantics, i.e., the absence of a measure of task-oriented source information. To address this problem, we propose a task-oriented concept for source information validity and source compression algorithms. Specifically, we consider semantic communication models in single- and multi-source systems and use the semantic context as well as correlation among sources to reduce redundant transmissions of information. We also study the relationship of code rates between Shannon, distributed, and semantic source coding, detailing the properties of the task and source under equal rates. Simulation results using public datasets demonstrate the validity and feasibility of the proposed semantic source coding algorithm across multiple communication tasks.