Background Knowledge Aware Semantic Coding Model Selection
Fangzhou Zhao, Yao Sun, Runze Cheng, Muhammad Ali Imran · 2022
Semantic communication is deemed to break Shannon channel capacity by transmitting extracted semantics rather than all binary bits. One critical challenge in semantic communication system is how to select a matching semantic coding model (SCM) in light of complicated source information, diversified user background knowledge (BK) and dynamic wireless channel. In this paper, we mathematically model the relationship among different BKs by using graph theory, and introduce a metric to evaluate SCMs performance as per BK relationships. Then, we propose a Background knowledge Aware SCM SElection (BASE) scheme, where a deep learning algorithm is exploited to accurately predict SCM performance in context of the modeled BK, guiding the SCM selection. Numerical simulation results show that the BASE has superiorities in information recovery accuracy along with the probability of selecting the optimal SCM when compared with other benchmarks.