【摘要】 提出一种Attention-Autoencoder模型并应用于系统中.将Attention模型与信道统计信息的相关先验知识结合,在接收端引入注意力(Attention)模块筛选有效数据从而增加网络提取特征的能力.基于Bellhop水声信道模型开展了不同信道环境的仿真实验验证,在测试集与训练集相差较大时,Attention-Autoencoder模型的水声通信系统具有较低误码率,具有较好的环境适应性.更多还原

【Abstract】 Autoencoder(AE) is globally optimized through two neural network modules at the transmitter and receiver and uses an end-to-end training method to improve the reliability of the communication system. However, the existing AE model is more sensitive to the underwater acoustic channel and has poor environmental adaptability. To improve the adaptability of the channel environment of the underwater acoustic communication system, the Attention-Autoencoder model is applied to the system. At the receiver, the attention module is introduced to filter effective data to increase the ability of the network to extract features. Combining the Attention model with relevant prior knowledge of channel statistical information, when the testing and the training are quite different, the underwater acoustic communication system based on this model still has a low bit error rate. Under the Bellhop hydroacoustic channel model, simulation experiments are carried out to verify different channel environments. The results show that the underwater acoustic communication system based on the proposed network model has the advantage of environmental adaptability.更多还原