DOA estimation algorithm based on complex-valued neural networks
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
维普
万方数据
传统模型驱动的波达方向(direction of arrival,DOA)估计算法性能受限于有限的信号特征、快拍数、信噪比、信杂比等因素,在低信噪比、快拍数少的极端情况下,性能较差.为克服上述问题,提高在极端条件下的估计精度,文章提出基于深度复数神经网络(complex-valued neural networks,CVNN)的单快拍DOA估计算法,构建深度复数神经网络模型,学习原始带噪信号与理想无噪复信号之间的映射关系,进而实现噪声抑制和期望信号特征增强的目的,提高DOA估计精度.仿真实验结果表明,经CVNN增强后,数据的等效信噪比约提高了 1 dB,等效快拍数提高了 3,该文所提算法相较于已有的多种物理驱动算法而言,具有更高的估计精度和泛化性.
The performance of traditional model-driven direction of arrival(DOA)estimation algorithm is limited by the finite signal characteristics,number of snapshots,signal-to-noise ratio(SNR),sig-nal-to-heterodyne ratio and other factors,and the algorithm performance is poor in the extreme cases of low SNR and few snapshots.To overcome the above problems and improve the estimation accuracy under extreme conditions,this paper proposes a single snapshot DOA estimation algorithm based on complex-valued neural networks(CVNN),which constructs a deep complex network model to learn the mapping relationship between the original noisy signal and the ideal noise-free complex signal,and then achieves noise suppression and desired signal feature enhancement.The proposed algorithm is used to improve the accuracy of DOA estimation.Simulation results show that after CVNN enhance-ment,the equivalent SNR of the data is improved by about 1 dB,and the equivalent number of snap-shots is improved by 3,the proposed algorithm has higher estimation accuracy and generalization than the existing multiple algorithms driven by physics.
direction of arrival(DOA)estimationcomplex-valued neural networks(CVNN)data drivenmodel driven
NETL
NSTL
维普
万方数据
