Experimental simulations of underwater acoustic semantic communication
[Objective]Underwater acoustic channels are normally limited by narrow bandwidth and are interfered by complicated environmental noise.These factors make high-efficiency and reliable underwater acoustic communication difficult to maintain.Semantic communication is characterized by microvolume representation and robust transmission.It could potentially improve underwater acoustic communication performance by overcoming narrow bandwidth limitation and complicated environmental noise interference.To validate the feasibility of revolutionizing underwater acoustic communication through semantic communication,an experimental simulation scheme of underwater acoustic semantic communication is developed.[Methods]This paper investigates and experimentally simulates data transmission through underwater acoustic channels based on semantic communication.The main process of the underwater acoustic semantic communication simulation scheme is presented as follows.Texts are the information source to be transmitted during the simulation.In the transmitter part,entities and relationships are extracted from texts using machine learning based on named entity recognition and relation extraction techniques.They form knowledge graphs.These knowledge graphs are concise semantic representations of the texts.In this process,knowledge graphs emerge as microvolume information carrier.This representation method does not require a large bandwidth of underwater acoustic channels to transmit data.In the receiver part,the texts are reconstructed from the knowledge graphs using machine learning based synonymous rephrasing techniques that are robust in noisy environments.The reconstructed texts are consistent with the originally transmitted texts at the semantic level.Therefore,efficient and reliable data transmission through underwater acoustic channels is possible in terms of semantically representing,transmitting,and reconstructing the data.To conduct the experimental simulations,a dataset comprising annotated Chinese descriptive texts for 520 underwater scenarios is established.Evaluation metrics include the number of transmitted bits,the bilingual evaluation understudy(BLEU)score,which measures word-level similarity,and the semantic similarity score,which measures sentence-level similarity.These metrics are used to evaluate the experimental results.[Results]This paper compares underwater acoustic semantic communication experimental simulations with traditional underwater acoustic communication experimental simulations.The experimental results are analyzed as follows:1)With an increase in the quantity of texts,the underwater acoustic semantic communication scheme consistently transmits fewer bits than traditional underwater acoustic communication scheme.2)For the same transmitted texts,the BLEU score and semantic similarity score of the underwater acoustic semantic communication scheme remain generally stable within a low signal-to-noise ratio(SNR)range and demonstrate superior overall performance.[Conclusions]The experimental simulation results confirm that the underwater acoustic semantic communication scheme exhibits better data compression and stronger robustness in data transmission than the traditional underwater acoustic communication schemes.It extracts semantic information from the data,significantly reducing the transmitted data volume and decreasing the channel bandwidth requirements.In addition,data are reconstructed at the receiver part based on machine learning.This capability achieves robust data reconstruction that reduces the influence of complicated environmental noise.The underwater acoustic semantic communication scheme possesses technical advantages such as microvolume expression and reliable transmission,thus forming a new underwater acoustic communication baseline that overcomes the difficulties of narrow bandwidth and low SNR raised by underwater acoustic channels.Therefore,it offers a new strategy for revolutionizing underwater acoustic communication.
NETL
NSTL
万方数据
