Scholarlink

<p>We propose a complex-valued neural network (CVNN) classification model based on generative probability waves for direction-of-arrival (DOA) estimation. Unlike existing CVNN approaches, our method employs linear regression instead of logistic regression, generating a continuous complex signal and analyzing its frequency spectrum to estimate unknown source DOAs. This avoids information loss caused by converting complex to real-valued representations. Due to the generative mechanism of probability waves, the proposed model inherently adapts to varying DOA angle ranges and resolution requirements without modifying the output layer dimensionality of the neural network. Additionally, we integrate time-step embedding into the model and incorporate the derivative relationship of probability waves with respect to time steps into the loss function, imposing stronger constraints on the generated results. Simulations demonstrate that the proposed method outperforms existing approaches, achieving higher estimation accuracy and improved robustness in various scenarios.</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

DOA Estimation Using Complex-Valued Neural Networks with Generative Probability Wave

  • 期刊论文

  • 发表日期：

    2025年9月24日

查看全文
  • Wenjie Xu,
  • Shichao Yi,
  • Ziyan Zhang

摘要

We propose a complex-valued neural network (CVNN) classification model based on generative probability waves for direction-of-arrival (DOA) estimation. Unlike existing CVNN approaches, our method employs linear regression instead of logistic regression, generating a continuous complex signal and analyzing its frequency spectrum to estimate unknown source DOAs. This avoids information loss caused by converting complex to real-valued representations. Due to the generative mechanism of probability waves, the proposed model inherently adapts to varying DOA angle ranges and resolution requirements without modifying the output layer dimensionality of the neural network. Additionally, we integrate time-step embedding into the model and incorporate the derivative relationship of probability waves with respect to time steps into the loss function, imposing stronger constraints on the generated results. Simulations demonstrate that the proposed method outperforms existing approaches, achieving higher estimation accuracy and improved robustness in various scenarios.