Environment Semantics Aided Underwater Wireless Optical Semantic Communication

Underwater wireless optical communication (UWOC) has emerged as a key technology for next-generation underwater wireless data transmission. Semantic communication is regarded as a promising communication paradigm to improve the efficiency and robustness of UWOC systems. However, existing underwater wireless optical semantic communication (UWOSC) system lacks the capability to adapt the code rate to different channel states, leading to intelligence inadequacy and coding inefficiency. In this paper, we introduce environment semantics into the UWOSC system for the first time, and propose an environment semantics aided underwater wireless optical semantic communication (ESA-UWOSC) framework. Based on this framework, we develop a system leveraging computer vision and deep learning for adaptive rate transmission. The ESA-UWOSC system is realized by a channel state perception mechanism (CSPM) for extracting environment semantics from the channel environment image to obtain the channel gain, a link network (LinkNet) for enabling variable code rate transmission of source semantic information, and a prediction network (PreNet) for predicting the optimal code length that satisfies the transmission task requirement. The combined operation of these three modules enables the ESA-UWOSC system to actively perceive channel state and perform adaptive rate transmission accordingly. In a series of experiments conducted on an emulated UWOC experimental platform, the results show that the ESA-UWOSC system outperforms the baseline scheme, achieving superior performance with lower bandwidth consumption, especially in harsh channel environments. Moreover, the results also demonstrate that the ESA-UWOSC system can achieve adaptive rate transmission based on the perceived channel state to meet transmission task requirement.