宽带/超宽带紧耦合天线阵的优化设计研究

Tightly coupled antenna array is wideband or ultra-wideband phased array antenna the future direction of development. To solve the problem of wideband or ultra-wideband tightly coupled antenna array parameters optimization and design efficiency to carry out efficient design method for wideband or ultra-wideband antenna array and coupling pattern synthesis research in this project. Through the analysis of based mode transmission line model and the higher order modes of transmission line with different propagation constant and characteristic impedance, the equivalent circuit model of tightly coupled antenna array can be established accurately. Then, Differential evolution (DE) algorithm is collaborative with the Gaussian Process (GP) machine learning. Machine learning process by using the Gauss nonparametric inference model to get an agent instead of numerical calculation means the evaluation of adaptability, improve the efficiency of the algorithm. Finally, DE-GP collaborative algorithm can be used to optimize the parameters of the equivalent circuit model of tightly coupled antenna array, and the collaborative algorithm can search sub-array row plan and sub array level weighted value, which exceeded tightly coupled antenna array pattern synthesis technology. In order to meet the demand of complex array design, the project will realize that the algorithm is accelerated by the aid of CUDA platform using CPU combined with GPU. The above research will lay the foundation for the development of the next generation phased array antenna.

紧耦合天线阵是未来宽带/超宽带相控阵天线的发展方向。本项目拟针对紧耦合天线阵的参数优化以及设计效率等问题开展宽带/超宽带紧耦合天线阵的高效设计方法及方向图综合研究。通过对紧耦合天线阵基模传输线模型和具有不同传播常数和特性阻抗的高次模传输线模型的分析，研究并建立紧耦合天线等效电路模型；然后，将差分（DE）进化算法协同高斯过程（GP）机器学习，利用高斯过程机器学习的非参数性推断得到代理模型以代替数值计算手段进行适应度评价，提高算法效率；最后，采用DE-GP协同算法对紧耦合天线阵的等效电路模型进行参数优化，并通过协同算法搜索子阵划分方案和子阵级加权值突破紧耦合天线阵的方向图综合技术问题；为了满足复杂阵列设计的需求，项目拟基于CUDA平台利用CPU结合GPU实现算法并行加速。以上研究可为下一代新体制相控阵天线的发展奠定基础。

课题组已按时完成了计划书中的研究内容，实现了研究目标。首先，本项目研究了紧耦合相控阵天线的宽带/超宽带性能并采用高次模传输线法分析了紧耦合相控阵天线的等效电路模型。然后，将DE算法和高斯过程机器学习算法进行协同，利用高斯机器学习算法中非凭经验对样本的训练数据获得满意的高斯过程代理模型，通过置信区间来对模型预测输出的不确定性进行评价，并将协同算法应用于紧耦合相控阵的优化设计中，提高了设计效率。其次，基于子阵的划分思想运用DE-GP全局优化算法对紧耦合阵列天线的方向图综合进行研究。最后，为了满足复杂阵列设计的需求，使用CPU+GPU异构体系对DE-GP协同算法进行并行加速，利用粗细嵌套模型对线程进行合理分配，提高并行执行效率。通过以上研究，建立了具有优越性能宽带/超宽带紧耦合天线阵的高效优化设计体系，达到了预期效果。