Intelligent Task-Oriented Semantic Communications: Theory, Technology and Challenges
LIU Chuanhong,GUO Caili,YANG Yang,CHEN Jiujiu,ZHU Meiyi,SUN Lu'nan
DOI: https://doi.org/10.11959/j.issn.1000−436x.2022117
2022-01-01
Abstract:Objectives: In the future, intelligent interconnection of all things, such as machine-to-machine and human-to-machine, poses challenges to traditional communication methods. The semantic communication method that extracts semantic information from source information and transmits them provides a novel solution for the sixth generation (6G) communication system. However, there are challenges in how to measure semantic information and how to achieve optimal semantic codec.This paper reviews the existing works related to semantic communication,and proposes a semantic communication method and framework for intelligent tasks,paving the way for further development of semantic communication. Methods: Firstly, the development history and research status of semantic communication are reviewed, the two bottleneck problems faced by semantic communication are analyzed and summarized, and a semantic communication method oriented to intelligent tasks is proposed. Aiming at the difficulty of semantic entropy,this paper defines the smallest basic unit of semantic message as semantic element,introduces fuzzy mathematics theory to describe the fuzzy degree of semantic understanding, and gives the calculation expression of semantic information entropy. Then, based on the information bottleneck theory, this paper proposes a semantic information coding scheme and a semantic channel joint coding scheme,respectively considering whether the receiver needs to reconstruct the original source. Furthermore, from the perspective of neural network interpretability,an interpretability-based semantic encoding method is proposed.Finally, a semantic communication platform for intelligent tasks is built based on software and hardware such as USRP and LabView,and the performance of the proposed algorithm is verified. Results:In the communication scenario where the source needs to be reconstructed,the semantic communication method proposed in this paper can greatly improve the compression ratio of the source data and greatly reduce the amount of transmitted data.Under the same compression ratio, the performance of the receiver to perform subsequent intelligent tasks can be improved,and the performance of source reconstruction can be improved at the same time.In scenarios where there is no need to reconstruct the source,the semantic communication method can better accomplish intelligent tasks with a large compression ratio.This is because semantic communication transmits the semantic information of the image instead of all the data of the image,which greatly reduces its bandwidth requirements,and the bandwidth utilization rate of semantic communication is 100 times higher than that of traditional communication methods. In addition, the anti-noise performance of the semantic communication method is much better than that of the traditional communication method, because the data transmitted by the semantic communication method retains the semantic features of the image,and the influence of channel noise is considered during model training, which makes the performance of intelligent tasks better and makes the communication system more robust. The semantic communication method greatly reduces the amount of data transmitted, so the transmission delay is significantly reduced under the same bandwidth resources.In addition,since image reconstruction is not required,the processing load of software and hardware is reduced, and the processing delay is also reduced. Therefore, the scheme proposed in this paper can greatly reduce the delay of end-to-end intelligent tasks while ensuring high-precision classification performance. Conclusions: Compared with traditional communication methods, the semantic communication method oriented to intelligent tasks has obvious advantages,which can greatly reduce the amount of transmitted data and improve the performance of intelligent tasks at the receiving end. Therefore, semantic communication will continue to maintain the trend of rapid development. However,there are still a lot of basic concepts and basic problems in semantic communication that need to be further discussed and improved,such as the basic theory of semantic information,the unified architecture of semantic communication,and the resource allocation strategy in semantic communication. Research is of great significance to promoting technological innovation and breakthroughs in the 6G era,and academic colleagues need to jointly promote the realization.