Scholarlink

<p>Direction-of-arrival (DOA) estimation is crucial in various applications. However, noise interference significantly impacts DOA estimation accuracy by altering signal strength, introducing distortion, or obstructing the target. The advantages of deep learning in array signal processing mainly lie in its adaptive learning ability, ability to handle nonlinear relationships, and end-to-end learning characteristics. A novel deep learning-based DOA estimation algorithm is proposed. To address the adverse impact of noise on DOA estimation accuracy, the proposed algorithm utilizes a bidirectional long short-term memory network to suppress noise present in the covariance matrix of array output signals. Achieving high-precision DOA estimation by using subspace technology typically necessitates prior knowledge of the number of sources. To estimate the number of sources from the denoised covariance matrix, the proposed algorithm also develops a gated recurrent unit network. Subsequently, the subspace technology is used to achieve DOA estimation. Experimental results show the effectiveness of the proposed algorithm in noise suppression and enhancement of DOA estimation accuracy.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Direction finding algorithm based on deep learning via subspace technology

  • 期刊论文

  • 发表日期：

    2025年5月12日

查看全文
  • Ruiyan Cai,
  • Quan Tian,
  • Gongrun Qiu

摘要

Direction-of-arrival (DOA) estimation is crucial in various applications. However, noise interference significantly impacts DOA estimation accuracy by altering signal strength, introducing distortion, or obstructing the target. The advantages of deep learning in array signal processing mainly lie in its adaptive learning ability, ability to handle nonlinear relationships, and end-to-end learning characteristics. A novel deep learning-based DOA estimation algorithm is proposed. To address the adverse impact of noise on DOA estimation accuracy, the proposed algorithm utilizes a bidirectional long short-term memory network to suppress noise present in the covariance matrix of array output signals. Achieving high-precision DOA estimation by using subspace technology typically necessitates prior knowledge of the number of sources. To estimate the number of sources from the denoised covariance matrix, the proposed algorithm also develops a gated recurrent unit network. Subsequently, the subspace technology is used to achieve DOA estimation. Experimental results show the effectiveness of the proposed algorithm in noise suppression and enhancement of DOA estimation accuracy.