| The concept of semantic communication has been proposed as early as when Shannon established information theory.However,in the past few decades,research on communication has mainly focused on how to transmit information accurately and efficiently,pursuing ”accurate”communication rather than ”expressive” communication.In the new era of ”Intelligent Internet of Everything”,people have new needs for communication,and research on semantic communication has gradually been put on the agenda.The advent of the era of artificial intelligence(AI)provides an opportunity for the study of semantic communication.Relying on the rapid development of natural language processing(Natural Language Processing,NLP)and deep learning(Deep Learning,DL)technology,the existing semantic communication has been able to successfully extract the semantic features of the transmitted information,and realize the information transmission based on semantic features.This paper focuses on the end-to-end semantic communication system,starting from the two perspectives of semantic encoding and decoding and semantic transmission,It provides a new direction for the existing semantic communication system.First of all,in terms of semantic coding and decoding,the existing semantic coding and decoding schemes based on a fixed neural network architecture lack flexibility,and all transmission scenarios will be allocated the same computing resources,ignoring the semantic differences between information and the signal-to-noise impact of transmission channels.In response to this phenomenon,this paper proposes a semantic encoding and decoding scheme based on Universal Transformer(UT).With the help of UT’s adaptive computing mechanism,the semantic encoding and decoding process will be affected by confidence.Different semantic features and channel signal noise will lead to differences in the recursion depth of the semantic network,thereby realizing the adaptive allocation of computing resources and improving the semantic system.overall transmission efficiency.In order to expand the overall practicability of the current semantic communication,the semantic transmission scheme is also a part that has to be considered.Different from the classical communication transmission based on statistical probability,for the semantic transmission based on logical probability,the semantic characteristics of the transmitted information are also elements that need to be considered in the transmission scheme.Different from the existing transmission schemes that only consider the channel Signal-Noise Ratio(SNR),the transmission scheme proposed in this paper is more based on the comprehensive consideration of semantic features and channel SNR.By combining with traditional communication systems,this paper proposes transmission schemes such as Hybrid Automatic Repeat re Quest(HARQ),rate control,and adaptive denoising,which are suitable for semantic communication.It effectively expands the scope of application of the semantic communication system,enables it to flexibly respond to different communication situations,and reduces the required communication resources while ensuring the accuracy of semantic transmission.To sum up,as far as the semantic coding scheme is concerned,this paper proposes a new semantic coding scheme,that is,a UT-based semantic codec scheme.By comprehensively considering different communication scenarios,and adaptively adjusting the recursive depth of the semantic network,it ensures While transmitting the accuracy rate,the waste of computing resources is avoided.As far as the semantic transmission scheme is concerned,this paper proposes three transmission schemes suitable for semantic communication systems,namely HARQ,rate control,and adaptive denoising schemes.The combination of features and new features effectively improves the transmission efficiency of the semantic system,provides a new option for the existing end-to-end semantic communication mode,and further expands the application range of semantic communication. |
PDF Full Text Request