Scholarlink

<p>Noise introduces inaccuracies in channel state information (CSI), undermining the reliability and efficiency of wireless systems. Errors in CSI impede decision-making, leading to misallocated resources and ultimately diminishing both reliability and data rates. Alternative approaches to managing this noise are impractical to implement for downlink channels within the scope of current 5G networks. We propose a practical method for denoising downlink CSI at the base station using time-series analysis. A synthetic dataset is used to train and evaluate a series of traditional and machine learning methods. Among the many models tested, the ensemble learning model that averaged the predictions of random forest, XGBoost, CatBoost, and support vector regressor models emerged as the most effective overall, achieving an average <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {R}^{2}\)</EquationSource> </InlineEquation> of 0.57 across all CSI metrics and reaching a maximum of 0.89 on specific key CSI values. Our approach shows significant promise for improving beamforming accuracy and real-time channel quality estimation in existing 5G deployments.</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

Denoising 5G limited channel state information

  • 期刊论文

  • 发表日期：

    2026年1月20日

查看全文
  • Ben Earle,
  • Ala’a Al-Habashna,
  • Gabriel Wainer,
  • Xingliang Li,
  • Guoqiang Xue

摘要

Noise introduces inaccuracies in channel state information (CSI), undermining the reliability and efficiency of wireless systems. Errors in CSI impede decision-making, leading to misallocated resources and ultimately diminishing both reliability and data rates. Alternative approaches to managing this noise are impractical to implement for downlink channels within the scope of current 5G networks. We propose a practical method for denoising downlink CSI at the base station using time-series analysis. A synthetic dataset is used to train and evaluate a series of traditional and machine learning methods. Among the many models tested, the ensemble learning model that averaged the predictions of random forest, XGBoost, CatBoost, and support vector regressor models emerged as the most effective overall, achieving an average \(\hbox {R}^{2}\) of 0.57 across all CSI metrics and reaching a maximum of 0.89 on specific key CSI values. Our approach shows significant promise for improving beamforming accuracy and real-time channel quality estimation in existing 5G deployments.