Scholarlink

<p>Receiving signals from global navigation satellite systems (GNSS) to offer users convenient and cost-effective positioning and navigation services has become an essential feature of smartphones. However, due to the low cost of GNSS modules and the challenges posed by complex urban environments, GNSS positioning solutions on smartphones often experience significant performance degradation from noise and multipath effects in pseudorange measurements. Both physical and statistical models struggle to accurately predict these disturbances. To overcome this limitation, we propose a hybrid neural network that integrates kernel extreme learning machine (KELM) with particle swarm optimization (PSO) to predict pseudorange errors using GNSS measurement features, including carrier-to-noise density ratio, elevation angle, Doppler, pseudorange residual, and pseudorange uncertainty. In our experiments, the proposed model is compared with traditional machine learning algorithms. When tested on real dynamic data collected from various environments, the proposed method achieves pseudorange error predictions with an average deviation of 2.80&#xa0;m. Using the pseudorange errors predicted by the proposed method to correct pseudorange measurements in single point positioning (SPP) reduces the 3D positioning root mean square error (RMSE) from 23.62 to 5.14&#xa0;m in mild urban environment and from 24.08 to 7.66&#xa0;m in deep urban environment. This represents a significant improvement over Doppler smoothing pseudorange SPP, which achieves RMSEs of 9.01&#xa0;m and 14.18&#xa0;m under the same conditions, respectively.</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

Pseudorange error prediction and correction using PSO-KELM for smartphone GNSS urban positioning

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Xuewei Huang,
  • Zhiqiang Dai,
  • Xiangwei Zhu

摘要

Receiving signals from global navigation satellite systems (GNSS) to offer users convenient and cost-effective positioning and navigation services has become an essential feature of smartphones. However, due to the low cost of GNSS modules and the challenges posed by complex urban environments, GNSS positioning solutions on smartphones often experience significant performance degradation from noise and multipath effects in pseudorange measurements. Both physical and statistical models struggle to accurately predict these disturbances. To overcome this limitation, we propose a hybrid neural network that integrates kernel extreme learning machine (KELM) with particle swarm optimization (PSO) to predict pseudorange errors using GNSS measurement features, including carrier-to-noise density ratio, elevation angle, Doppler, pseudorange residual, and pseudorange uncertainty. In our experiments, the proposed model is compared with traditional machine learning algorithms. When tested on real dynamic data collected from various environments, the proposed method achieves pseudorange error predictions with an average deviation of 2.80 m. Using the pseudorange errors predicted by the proposed method to correct pseudorange measurements in single point positioning (SPP) reduces the 3D positioning root mean square error (RMSE) from 23.62 to 5.14 m in mild urban environment and from 24.08 to 7.66 m in deep urban environment. This represents a significant improvement over Doppler smoothing pseudorange SPP, which achieves RMSEs of 9.01 m and 14.18 m under the same conditions, respectively.