摘要
针对相干信源下的到达角估计分辨率低的问题,提出了基于广义回归神经网络(GRNN)模型的估计方法。建立协方差矩阵与相应目标角度的映射关系形成训练集,经过广义回归神经网络模型进行监督学习获得模型参数;分析比较模型传递函数中扩散系数对系统性能的影响,将不同径向基函数扩散系数的预测结果与训练标签之间的最小值选为扩散系数最优值,进一步提升了估计精度。仿真结果表明,采用GRNN的估计方法可在不牺牲有效阵元数的情况下,达到解相干估计的目的。该方法的估计误差在0.5°范围内的准确率达96.33%,高于多信号分类方法(MUSIC)估计和空间平滑估计方法,且均方误差值更小,结果更加精确。
Aiming at the low-resolution problem of DOA estimation for coherent sources,an estimation method based on the generalized regression neural network(GRNN)model was proposed.This method first establishes the mapping relationship between the covariance matrix and the corresponding target angle to form a training set,and obtains model parameters through supervised learning of the generalized regression neural network model.The influence of the diffusion coefficient in the model transfer function on the system performance was analyzed and compared.The minimum value between the prediction result of the diffusion coefficient and the training label was selected as the optimal value of the diffusion coefficient,which further improved the estimation accuracy.The simulation results show that the GRNN estimation method can achieve the purpose of decoherent estimation without sacrificing the number of effective array elements.The estimation accuracy rate of this method is 96.33%in the error range of 0.5°,which is higher than that of the MUSIC estimation and spatial smoothing estimation methods,and the mean square error is smaller,the result is more accurate.
作者
翁元博
徐湛
田露
职如昕
WENG Yuanbo;XU Zhan;TIAN Lu;ZHI Ruxin(School of Information and Communication Engineering,Beijing Information Science&Technology University,Beijing 100192,China)
出处
《北京信息科技大学学报（自然科学版）》
2021年第4期66-71,75,共7页
Journal of Beijing Information Science and Technology University
基金
北京市科技计划课题(Z191100001419001)
北京市优秀人才资助计划青年拔尖项目(2016000026833ZK08)
北京信息科技大学现代测控技术教育部重点实验室资助(KF20201123202)。
