.png){kind=icon}
.png){kind=icon}
Return
Semantic Communications Toward Graph Data
.png){kind=icon}
PRE
AI
DOI：10.1109/MWC.002.2400034.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In the development of communication systems, semantic communications (SemCom) represents a pivotal shift from conventional methods, underscoring the primacy of semantic understanding in data transmission. The burgeoning relevance of graph data in delineating intricate real-world interconnections, underscores the necessity of integrating it within SemCom frameworks. However, unlike traditional unstructured data, such as images and texts, the semantic information inherent in graph data is more abstract and less directly perceptible. This distinction poses a challenge to existing SemCom systems, of which the designs have not yet considered the transmission of such types of data and their associated semantic information. To address this problem, this article introduces a topology-aware graph semantic coding (TAGSC) framework. The framework emphasizes the critical need for understanding graph topology to effectively transmit and interpret graph semantic information. Within this conceptual framework, we propose the graph information bottleneck joint semantic-channel coding (GIB-JSCC) scheme, a bespoke instantiation of TAGSC tailored for the graph classification task in the context of SemCom. GIB-JSCC integrates the information bottleneck principle and the JSCC method. Particularly, it is engineered to filter out task-irrelevant graph information while capturing task-relevant ones. Our case study demonstrates GIB-JSCC's efficacy in improving SemCom's performance on the graph semantic information transmission, especially in surpassing existing methods notably in low SNR conditions. With an open ending, the article concludes by discussing future prospects and challenges in SemCom for graph data.
Keywords:
Semantics
Topology
Encoding
Task analysis
Decoding
Atoms
Codecs
Journal
IF:
11.5
Papers:
2.7K
Citations:
1.3W