基于注意力机制的LSTM时变水声信道深度学习预测

This paper investigates the channel prediction algorithm of the time-varying channels in underwater acoustic(UWA)communication systems using the long short-term memory(LSTM)model with the attention mechanism.AttLstmPreNet is a deep learning model that combines an attention mechanism with LSTM-type models to capture temporal information with different scales from historical UWA channels.The attention mechanism is used to capture sparsity in the time-delay scales and coherence in the gep-time scale under the LSTM framework.The soft attention mechanism is introduced before the LSTM to support the model to focus on the features of input sequences and help improve the learning capacity of the proposed model.The performance of the proposed model is validated using different simulation time-varying UWA channels.Compared with the adaptive channel predictors and the plain LSTM model,the proposed model is better in terms of channel prediction accuracy.

鞭毛/纤毛内在驱动机制启发的一体式管状机器人驱动器

鞭毛和纤毛独特的运动模式(如鞭毛的平面/螺旋波形式推进和纤毛的二维/三维不对称搏动)在众多生物的生命活动中起到至关重要的作用,这也启发了诸多仿生设计,尤其对于微型机器人系统。然而,与自然界中微生物能够从统一化的9+2轴丝生物结构中进化出多种运动模式不同的是,当前的仿生学仍然没有有效的工程策略去实现这样的智慧。在此,我们通过研究鞭毛和纤毛的内部结构及其内在驱动机制,推导出了一个统一的物理模型来描述微管弯曲及其所构建的宏观鞭毛/纤毛运动。基于该模型,我们进而提出了基于三通道的管状驱动概念,并相应地通过杆嵌入铸造工艺制造了一个三通道的管状驱动器。通过编程不同通道的驱动模式,这一管状驱动器不仅可以再现自然界中多样的二维及三维鞭毛/纤毛运动,还可以延展出更多的非对称纤毛搏动模式以实现低雷诺数下的有效推进。该研究加深了我们对微生物推进机制的理解,为仿生系统的设计提供了新灵感,并有望在广泛的工程领域中寻找到重要的应用场景。