基于深度学习的智能天线智能化增强方法研究

As the window of delivering information, antenna will carry the heavy task of sensing complicated environment and conducting direct information interaction for intelligent machines, so itself also needs to be highly intelligentalized. The concept of smart antenna was early proposed in 1990’s and it mainly adjusts the directional pattern of beams in an adaptive way by adjusting each element’s weight of signal in an antenna array so that the useful signal can be maximized while interference signal can be suppressed. However, such kind of antenna is not intelligentalized in fact although it was named “smart” antenna, and it only includes certain automation function instead. The standard of distinguishing intelligence from automation is to see if a machine or device can continuously enhance its capability by learning and evolving. This project plans to investigate the approach and method of realizing a real intelligentalization for smart antennas by making antenna’s hardware become soft and adopting the newest technology of artificial intelligence (AI). In hardware, the newest soft or liquid electronic materials will be used to build antenna array so that the degree of freedom and flexibility of its performance reconfiguration can be significantly enhanced by making its physical form become plasticized. In software, we will mainly focus on constructing the intelligentalization part, namely, we shall study the method of optimally reconfiguring antenna’s performance and making it possess self-evolution ability by aiming to the complicated, changeable, and possibly-uncertain characteristics of signals and adopting the deep learning model in the newest AI technology. The research result will be validated by computer modeling and simulation.

天线作为信息传递的窗口承载着未来智能机器对复杂环境感知与信息直接交互作用的重任，其自身也需要高度的智能化。智能天线的概念早在1990年代就已提出，主要通过调节阵列天线中各阵元信号权重而自适应调节波束的方向图，以使有用信号得到最大化，干扰信号得到抑制。然而，这种天线虽然名为智能天线但实际上并没有智能化，而只是自动化。区分智能化与自动化的标准是看机器或装置是否能够像人一样通过学习与进化不断增强其能力。本项目拟通过使天线硬件变软并采用最新人工智能技术研究使智能天线真正实现智能化的途径和方法。在硬件上拟采用最新的柔性或液体电子材料来制作天线，通过使天线的物理形态可塑化来提高天线性能重构的自由度和灵活性。在软件上着重构建其智能化部分，即针对外来复杂多变且具有不确定性的信号特征采用最新人工智能技术中的深度学习模型研究重构最优天线性能的方法并使其具有自我进化的能力。研究结果将通过计算机仿真手段加以验证。

本项目针对智能天线在当今军事电子对抗或无线通信中的重要作用，研究大幅度提升智能天线智能化程度的途径。机器智能化的特征是具有自感知、自学习、自进化的能力，这需要硬件和软件都能根据环境或任务的需要随时动态调整。传统上硬件一般是不可改变的，这大大限制了机器性能的可调控性与可重构性。本项目从探索使硬件软化的途径着手，提出采用液体金属来制作智能天线阵列单元以使整体智能天线物理形态可改变、可调控，大大增加天线性能可重构的自由度。经过一年时间的研究，本项目取得了以下重要成果：1）首次实现了两维（2D）液体金属共熔镓铟合金（EGaIn）表面的可调控性，通过氢氧化钠（NaOH）电解液来改变EGaIn表面张力从而调控其流动与收缩，使天线的工作频率可以在0.71GHz至4.03GHz范围内连续可调，最大增益达到3.48dBi。2）研究并设计了一款可调的频率选择吸收器，该吸收器由一个圆形腔体和3D打印的阶梯式衬底构成，通过电化学方法改变液体金属EGaIn的半径，吸收器的频率响应可连续重构。实验表明当EGaIn的半径在6-9mm范围时吸收器可达到大于25dB的带内吸收率和小于5dB的带外损耗。3）研究并设计了一款利用微流体和液体金属技术并通过3D打印制作的倒F型天线，可用作无线无源温度传感器。我们利用液体金属（水银）的热胀冷缩改变其物理形状、进而改变其谐振频率的原理，建立温度与谐振频率的对应关系，通过测量天线的谐振频率来测量温度，从而实现远距离无线监测温度。4）研究了天线设计中的两种优化算法，即基因算法（GA）和混沌粒子群优化算法（CPSOA）。针对人工智能算法中需要不断学习和优化、大样本训练收敛进程缓慢的特点，我们提出了能显著提高优化设计效率的改进型GA和自适应CPSOA，仿真结果表明前者效率可提高89%而后者可提高56%，而且后者有更低旁瓣和更深零点的辐射模式。上述成果的科学意义在于为2D液体金属阵列单元的设计与制作提供了理论指导并验证了其用来调控天线频率特性的可行性，为以后融入基于深度学习的人工智能技术建立了良好基础。