摘要
在火灾和地震等灾害场景中,由无人机等平台构成的空中应急通信系统应具备高可靠、低冗余的文本传输能力。语义通信作为一种新兴的通信范式,借助人工智能技术提取文本中的关键信息以降低冗余,并在恶劣信道条件下仍有良好的语义重构能力,为实现这一需求提供了新的技术路径。面向无人机网络场景,首先分析了文本语义传输的特点及需求,包括动态信道自适应、语义模型轻量化、语义重构一致性、语义优先级评估等,并总结了现有的研究方案及不足。其次,针对上行文本指令下发与下行态势数据回传这两类典型业务,设计了面向上行链路的单模态文本语义传输架构,以及面向下行链路的多模态语义传输及融合架构。再次,在所提语义传输架构基础上,提出了一种面向无人机网络的文本语义适变传输策略,引入多头注意力机制以捕捉长距离语义依赖,并采用最小化交叉熵与最大化互信息的联合优化策略,实现了在动态信道条件下的文本语义可靠传输。仿真结果表明,在信噪比(SNR)为0~5 dB时,所提方案的BLEU-4得分远高于传统Huffman+LDPC的分离式信源信道编码传输方案;在SNR为5~10 dB时,性能进一步提升,表明了所提文本语义传输方案在不同信道环境下均具有良好的的自适应性和语义重构能力。
In disaster scenarios such as fires and earthquakes,aerial emergency communication systems composed of unmanned aerial vehicles(UAVs)must deliver textual information with high reliability and minimal redundancy.Semantic communication,as an emerging communication paradigm,leverages artificial intelligence techniques to extract key information from text,thereby reducing redundancy,and demonstrates robust semantic reconstruction capabilities under harsh channel conditions,offering a new technical pathway to meet this demand.Targeting UAV networks,this study first analyzes the characteristics and requirements of textual semantic transmission,including dynamic channel adaptation,lightweight semantic models,semantic reconstruction consistency,and semantic priority assessment,and the existing approaches along with their limitations are summarized.Next,for two typical services of uplink textual command delivery and downlink situational data reporting,we design a unimodal text semantic transmission architecture for the uplink and a multimodal semantic transmission and fusion architecture for the downlink.Furthermore,based on the proposed semantic transmission architecture,a UAV-oriented adaptive textual semantic transmission strategy is proposed.Specifically,a multi-head attention mechanism is used to capture long-range semantic dependencies,and a joint optimization approach is adopted to simultaneously minimize the cross-entropy and maximize the mutual information,thereby ensuring robust semantic delivery over time-varying channels.Simulation results show that at a signal-to-noise ratio(SNR)of 0~5 dB,the proposed method a significantly higher BLEU-4 score than the conventional separated source-channel coding transmission scheme(Huffiman+LDPC).When the SNR is at 5~10 dB,the performance is further improved,demonstrating that the proposed text semantic transmission scheme exhibits strong adaptability and semantic reconstruction capability across different channel environments.
作者
王彦曈
张欣欣
刘宜明
张治
WANG Yantong;ZHANG Xinxin;LIU Yiming;ZHANG Zhi(State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China)
出处
《移动通信》
2025年第10期119-127,共9页
Mobile Communications
基金
国家自然科学基金“面向不均衡联邦学习的无线算力可信协同机理研究”(62471065)
北京市自然科学基金资助项目“复杂工业网络下面向语义通信的端边协同训推优化方法研究”(L251036)。
关键词
无人机网络
语义通信
文本传输
联合信源信道编码
UAV networks
semantic communication
text transmission
joint source-channel coding
