Scholarlink

<p>Accurate direction-of-arrival (DOA) estimation is crucial for applications like antenna array beamforming and wireless localization. Traditional DOA methods such as MUSIC, ESPRIT, and Capon often struggle with computational complexity and degraded performance in challenging environments. This paper provides a deep learning-based DOA estimator applied to real-time complex samples collected using an FMCOMMS2 and Zed Board SDR dual-channel receiver with Vivaldi Antenna spaced at half-wavelength. Measurements in the azimuth plane from 30°to 130°in 10°increments, generating a dataset for training a deep neural network (DNN). The proposed DNN architecture effectively maps the features of signals to DOA with a mean squared logarithmic error of <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(2.4583 \times 10^{-4}\)</EquationSource> </InlineEquation> and a validation loss of 1.01105 over 100 epochs. A correlation coefficient of <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(99.9\%\)</EquationSource> </InlineEquation> proves the model is reliable and has a high degree of 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

Neural Network-Based Direction of Arrival Estimation Using Real-Time Signal Capture from a Dual-Channel SDR Platform

  • 期刊论文

  • 发表日期：

    2026年3月6日

查看全文
  • Mahmoud A. Khalil,
  • Mohamed A. Abouelatta,
  • Hussein A. Atty,
  • Mohamed Mabrouk

摘要

Accurate direction-of-arrival (DOA) estimation is crucial for applications like antenna array beamforming and wireless localization. Traditional DOA methods such as MUSIC, ESPRIT, and Capon often struggle with computational complexity and degraded performance in challenging environments. This paper provides a deep learning-based DOA estimator applied to real-time complex samples collected using an FMCOMMS2 and Zed Board SDR dual-channel receiver with Vivaldi Antenna spaced at half-wavelength. Measurements in the azimuth plane from 30°to 130°in 10°increments, generating a dataset for training a deep neural network (DNN). The proposed DNN architecture effectively maps the features of signals to DOA with a mean squared logarithmic error of \(2.4583 \times 10^{-4}\) and a validation loss of 1.01105 over 100 epochs. A correlation coefficient of \(99.9\%\) proves the model is reliable and has a high degree of accuracy.