Scholarlink

<p>Global Navigation Satellite System (GNSS), widely used in city navigation and location-based service, suffers from the severe deterioration of accuracy and stability in harsh urban environments. Based on the Long Short-term Memory (LSTM) model, we propose a novel receiver clock modeling and prediction (RCMP) method to improve GNSS urban positioning, which takes into account the nonlinear relationship between receiver clock offsets, the up-coordinate and Time Dilution of Precision (TDOP). The proposed LSTM model is compared with the Long Expressive Memory (LEM), the gray (GM) and the quadratic polynomial (QP) models. Additionally, in the kinematic experiment where receiver clock predictions are more susceptible to noise interference, LSTM and LEM introduce the Gaussian Negative Log-Likelihood Loss as the training criterion to capture the predictive uncertainty, denoted as LSTM<sub>NLL</sub> and LEM<sub>NLL</sub>. LSTM achieves the best prediction accuracy, which is 10–45%, 7–34% and 36–83% better than LEM, GM and QP in the static experiment. In the kinematic experiment, LSTM<sub>NLL</sub> outperforms the others, improving by 8–18%, 2–79%, 4–71%, 9–70% and 26–78% compared to LSTM, LEM<sub>NLL</sub>, LEM, GM and QP, respectively. When applied to single point positioning (SPP), in the static experiment, LSTM improves the up direction by 10–64%, 39–54% and 48–88%. In the kinematic experiment, LSTM<sub>NLL</sub> enhances the up direction by 6–14%, 3–77%, 12–65%, 5–54% and 30–75%. The improvements to the east and north directions are also obvious, just a little slighter.</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

LSTM-based receiver clock modeling and prediction for GNSS urban positioning

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • Zhiqiang Dai,
  • Simei Sun,
  • Ting Xie,
  • Chunlei Zhai,
  • Zhigang Hu,
  • Xiangwei Zhu,
  • Shaofeng He,
  • Fan Feng

摘要

Global Navigation Satellite System (GNSS), widely used in city navigation and location-based service, suffers from the severe deterioration of accuracy and stability in harsh urban environments. Based on the Long Short-term Memory (LSTM) model, we propose a novel receiver clock modeling and prediction (RCMP) method to improve GNSS urban positioning, which takes into account the nonlinear relationship between receiver clock offsets, the up-coordinate and Time Dilution of Precision (TDOP). The proposed LSTM model is compared with the Long Expressive Memory (LEM), the gray (GM) and the quadratic polynomial (QP) models. Additionally, in the kinematic experiment where receiver clock predictions are more susceptible to noise interference, LSTM and LEM introduce the Gaussian Negative Log-Likelihood Loss as the training criterion to capture the predictive uncertainty, denoted as LSTMNLL and LEMNLL. LSTM achieves the best prediction accuracy, which is 10–45%, 7–34% and 36–83% better than LEM, GM and QP in the static experiment. In the kinematic experiment, LSTMNLL outperforms the others, improving by 8–18%, 2–79%, 4–71%, 9–70% and 26–78% compared to LSTM, LEMNLL, LEM, GM and QP, respectively. When applied to single point positioning (SPP), in the static experiment, LSTM improves the up direction by 10–64%, 39–54% and 48–88%. In the kinematic experiment, LSTMNLL enhances the up direction by 6–14%, 3–77%, 12–65%, 5–54% and 30–75%. The improvements to the east and north directions are also obvious, just a little slighter.