可信推断近场稀疏综合阵列三维毫米波成像  

Credible Inference of Near-field Sparse Array Synthesis for Three-dimensional Millimeter-wave Imagery

作　　者：杨磊 霍鑫 申瑞阳 宋昊 胡仲伟 YANG Lei;HUO Xin;SHEN Ruiyang;SONG Hao;HU Zhongwei(College of Electronic Information and Automation,Civil Aviation University of China,Tianjin 300300,China)

机构地区：[1]中国民航大学电子信息与自动化学院,天津300300

出　　处：《雷达学报（中英文）》2024年第5期1092-1108,共17页Journal of Radars

基　　金：国家自然科学基金(62271487);中央高校基本科研业务费(3122023PT04)。

摘　　要：考虑到主动式电扫描毫米波成像系统在实际应用中成像场景要求大,分辨率要求高,但毫米波的波长短,继而造成满足奈奎斯特采样定理的均匀阵列规模及馈电网络复杂度过高,面临着成像精度、成像速度和系统成本之间的矛盾。针对以上问题,该文提出了可信推断近场稀疏综合阵列算法(CBI-SAS),在全贝叶斯学习框架下,该算法基于贝叶斯推断对复激励权值进行稀疏优化,得到复激励权值的完全统计后验概率密度函数,从而利用其高阶统计信息得到复激励权值的最优值及其置信区间和置信度。在贝叶斯推断中,为了实现较少数量的阵元合成期望波束方向图,可通过对复值激励权值引入重尾的拉普拉斯稀疏先验。然而,由于先验概率模型与参考方向图数据模型非共轭,因此需对先验模型进行分层贝叶斯建模,从而保证得到的复激励权值完全后验分布具有闭合解析解。为了避免求解完全后验分布的高维积分,采用变分贝叶斯期望最大化方法计算复激励权值后验概率密度函数,实现复激励权值的可信推断。仿真模拟实验结果显示,相较于传统稀疏阵列合成方法,所提方法阵元稀疏度更低、归一化均方误差更小、匹配方向图精度更好。此外,基于设计的稀疏阵列采集近场一维电扫和二维平面全电扫实测回波数据后,利用改进三维时域算法进行三维重建,验证了所提CBI-SAS算法在保证成像结果的同时降低了系统复杂性的优势。Due to the short wavelength of millimeter-wave,active electrical scanning millimeter-wave imaging system requires large imaging scenarios and high resolutions in practical applications.These requirements lead to a large uniform array size and high complexity of the feed network that satisfies the Nyquist sampling theorem.Accordingly,the system faces contradictions among imaging accuracy,imaging speed,and system cost.To this end,a novel,Credible Bayesian Inference of near-field Sparse Array Synthesis(CBI-SAS)algorithm is proposed under the framework of sparse Bayesian learning.The algorithm optimizes the complexvalued excitation weights based on Bayesian inference in a sparse manner.Therefore,it obtains the full statistical posterior Probability Density Function(PDF)of these weights.This enables the algorithm to utilize higher-order statistical information to obtain the optimal values,confidence intervals,and confidence levels of the excitation weights.In Bayesian inference,to achieve a small number of array elements to synthesize the desired beam orientation pattern,a heavy-tailed Laplace sparse prior is introduced to the excitation weights.However,considering that the prior probability model is not conjugated with the reference pattern data probability,the prior model is encoded in a hierarchical Bayesian manner so that the full posterior distribution can be represented in closed-form solutions.To avoid the high-dimensional integral in the full posterior distribution,a variational Bayesian expectation maximization method is employed to calculate the posterior PDF of the excitation weights,enabling reliable Bayesian inference.Simulation results show that compared with conventional sparse array synthesis algorithms,the proposed algorithm achieves lower element sparsity,a smaller normalized mean square error,and higher accuracy for matching the desired directional pattern.In addition,based on the measured raw data from near-field 1D electrical scanning and 2D plane electrical scanning,an improved 3D time domain alg

关 键 词：毫米波成像 贝叶斯推断 稀疏阵列合成 分层贝叶斯 变分贝叶斯期望最大 

分 类 号：TN957[电子电信—信号与信息处理]