[1]王 鹏,姚爱琴,屈泽楷,等.基于深度展开与自适应融合的DOA 估计[J].机械与电子,2026,44(08):1-9.
　WANG Peng,YAO Aiqin,QU Zekai,et al.DOA Estimation Based on Depth Expansion and Adaptive Fusion[J].Machinery & Electronics,2026,44(08):1-9.
点击复制

基于深度展开与自适应融合的DOA 估计()
分享到：

《机械与电子》[ISSN:1001-2257/CN:52-1052/TH]

卷:
44
期数:
2026年08期
页码:
1-9
栏目:
研究与设计
出版日期:
2026-08-25

文章信息/Info

Title:
DOA Estimation Based on Depth Expansion and Adaptive Fusion
文章编号:
1001-2257(2026)08-0001-09
作者:
王 鹏; 姚爱琴; 屈泽楷; 申梦蕊
中北大学信息与通信工程学院,山西 太原 030051
Author(s):
WANG Peng; YAO Aiqin; QU Zekai; SHEN Mengrui
(School of Information and Communication,North University of China,Taiyuan 030051,China)
关键词:
DOA 估计; 深度展开; 自适应融合; 压缩感知
Keywords:
direction of arrival (DOA)estimation; deep expansion; adaptive fusion; compressed sensing
分类号:
TP391;TN911.7
文献标志码:
A
摘要:
针对低信噪比、少快拍及阵列位置误差等复杂条件下传统DOA 算法鲁棒性不足的问题,提出一种基于深度展开与自适应加权的DOA 网络DUAF-Net。该网络基于双分支网络结构,第一分支为使用注意力机制的DOA 聚焦网络;第二分支为压缩感知ISTA 算法的深度展开模块,用于引入稀疏先验约束。模型从2个数据支路学习,2个分支的输出经拼接后输入全连接层,融合得到最终的DOA 估计结果。此外,针对网络特性,设计了两阶段训练策略:首先在理想条件下对完整网络进行预训练,随后在低信噪比、少快拍及阵列位置误差条件下联合微调,以增强网络对恶劣环境的适应能力。仿真实验表明,在信源数为1~5、信噪比为-10~20 dB、快拍数为20~200的均匀线阵环境下,DUAF-Net在-10~20 dB区间的平均均方根百分比误差(RMSPE)为2.06×10-2,估计准确率在低信源数场景下达到98.2%,性能显著优于MUSIC、LISTANet等对比方法,为复杂电磁环境下高精度DOA 估计方法提供了可行方案。
Abstract:
Conventional direction-of-arrival (DOA) estimation algorithms suffer from insufficient robustness under adverse conditions such as low signal-to-noise ratio (SNR),limited snapshots,and array position errors.To address this issue,we propose DUAF-Net,a deep unfolding network with adaptive weighting for DOA estimation.The network adopts a dual-branch architecture:the first branch is a DOA focusing network equipped with an attention mechanism,while the second branch is a deep unfolding module that implements the iterative shrinkage-thresholding algorithm (ISTA) for compressed sensing, thereby incorporating sparse prior constraints.The model learns from two data pathways,and the outputs of the two branches are concatenated and fed into a fully connected layer to produce the final fused DOA estimate.Additionally,a two-stage training strategy is implemented to accommodate the network characteristics: first,the entire network is pre-trained under ideal conditions,and then jointly fine tuned under low SNR,few snapshots,and array position error scenarios to enhance adaptability to harsh environments. Simulation experiments demonstrate that,for a uniform linear array with 1—5 sources,SNR ranging from -10—20 dB,and snapshot numbers from 20—200,DUAF-Net achieves an average root mean square percentage error (RMSPE) of 2.06×10-2over the -10—20 dB range,and attains an estimation accuracy of 98.2% in low source number scenarios.The proposed method significantly out performs competing methods such as MUSIC and LISTANet,offering a feasible solution for high-precision DOA estimation in complex electromagnetic environments.

参考文献/References:

[1] 李俊霞,王欣,黄高见,等.无源定位技术发展及其展望[J].无线电工程,2024,54(8):1825-1846. [2] 白毓杰,郑桂妹,宋玉伟,等.基于 CNN-NAFNet级联网络的低信噪比 DOA 估计方法研究[J].信号处理,2026,42(4):491-505. [3] 陈志鹏,肖长诗,李昊鑫,等.基于奇异值分解与零空间相结合的多信源方向估计方法[J/OL].声学技术,2026-04-29.https:∥doi.org/10.16300/j.cnki.1000 3630.24123001. [4] 熊伟华,章勇,张雪梅,等.基于轻量化级联神经网络的信号 DOA 估计[J].电子测量技术,2026,49(5):156-167. [5] 王安义,马晨晨,王裕旗.基于快速傅里叶变换的长时间时变 DOA 精确估计[J].现代雷达,2026,48(2):51-57. [6] 薛林,陈鹏,吕德利.基于稀疏重构的高精度宽带 DOA估计方法[J].现代雷达,2026,48(1):62-67. [7] 王军,武子涵,周广佼,等.基于角度间隔分离学习的相干 DOA 估计方法[J].系统工程与电子技术,2025,47(10):3188-3198. [8] 刘其有,何瑞,施伟.基于深度学习的波达方向估计方法综述[J].中国电子科学研究院学报,2025,20(1):1-9. [9] 刘振,苏晓龙,师俊朋,等.稀疏阵列波达方向估计研究进展[J].信息对抗技术,2023,2(增刊1):1-15. [10] Papageorgiou G K,Sellathurai M,Eldar Y C.Deep networks for direction-of-arrival estimation in low SNR[J].IEEE Transactions on Signal Processing,2021,69:3714-3729. [11] Xu Qian,Gao Yulong,Zhang Ruoyu,et al.Deep learning-based doa estimation via grid search approach for unknown and variable number of signals[J].IEEE Transactions on Cognitive Communications and Networking,2026,12:2708-2723. [12] Merkofer J P,Revach G,Shlezinger N,et al.DA-MUSIC:data-driven DoA estimation via deep augmented MUSIC algorithm[J].IEEE Transactions on Vehicular Technology,2023,73(2):2771-2785. [13] Niu Xinyi,Su Xiaolong,He Lida,et al.Direction of arrival (DOA) estimation using a deep unfolded learned iterative shrinkage thresholding algorithm (LISTA) network in a non-uniform metasurface[J].Remote Sensing,2025,17(7):1253. [14] Zhu Hangui,Chen Xixi,Ma Teng,et al.Deep unfolded amplitude-phase error self-calibration network for DOA estimation[J].Journal of Systems Engineering and Electronics,2024,36(2):353-361. [15] Xu Wenjie,Yi Shichao.DACL Net:a dual branch attention-based CNN-LSTM network for DOA estimation[ J].Sensors,2026,26(2):743. [16] Zhang Zhenshan,Xu Wenjie,Zou Haitao,et al.A GAN-CNN fusion framework for deep learning-based DOA estimation in low-SNR environments[J].Sensors, 2026,26(5):1676. [17] 中国人民解放军空军工程大学.基于深度展开网络的 无网格波达方向估计方法及系统:CN202511115996.5[P].2025 11-11. [18] 杭州电子科技大学.基于块稀疏BCS深度展开的复值解卷积DOA估计方法:202511667330.0[P].2026-02-03. [19] Jiang Linqiang,Tang Tao,Wu Zhidong,et al.Fusion based off-grid direction of arrival estimation via deep learning with array imperfection[J].IEEE Transactions on Vehicular Technology,2026,75:45-56. [20] Zhou Bo,Xu Kaijie,Quan Yinghui,et al.A deep learning based supervised transfer learning framework for DOA estimation with array imperfections[PP/ OL].arXiv (2025-04-18)[2026-07-01].https:∥ arxiv.org/abs/2504.13394.arXiv:2504.13394.

备注/Memo

备注/Memo:
收稿日期:2026-06-15 基金项目:山西省基础研究计划资助项目 (202403021212022);山西省高等学校科技创新项目(2024L191) 作者简介:王 鹏 (2003-),男,山西临汾人,硕士研究生,研究方向为软件无线电定位技术;姚爱琴 (1969-),女,山西晋中人,博士,副教授,硕士研究生导师,研究方向为频谱压缩感知技术、物联网系统等,通信作者,E-mail:86093688@qq.com;屈泽楷(2001-),男,山西太原人,硕士研究生,研究方向为信号调制识别;申梦蕊 (2002-),女,河南南阳人,硕士研究生,研究方向为红外小目标检测。
更新日期/Last Update: 2026-08-28