Scholarlink

<p>Accurate channel estimation in dynamic wireless environments is a critical challenge for emerging 5G and 6G systems, where multipath components (MPCs) vary rapidly due to human mobility, obstacles, and frequent LoS-to-NLoS transitions. Traditional clustering techniques such as K-means or Gaussian mixture models provide useful statistical structures but are limited by fixed cluster assumptions, sensitivity to initialization, and poor adaptability under non-stationary conditions. This research proposes a novel deep learning framework that integrates adaptive, power-aware differentiable clustering with neural channel estimation to overcome these limitations. The method introduces a differentiable clustering layer that assigns MPCs to latent prototypes using power-weighted soft responsibilities, ensuring that high-power paths dominate cluster formation while weaker ones are treated proportionally. Prototypes are updated online with exponential moving averages and forgetting factors, enabling continual adaptation to time-varying environments. Cluster summaries, including prototype means, spreads, and power distributions, are then used as conditioning signals for a neural estimator via cross-attention and feature-wise modulation. The entire model is trained end-to-end to minimize channel estimation error while enforcing compact, stable, and interpretable clusters. Experimental validation with sub-6&#xa0;GHz USRP measurements in indoor scenarios demonstrates improvements of 18–22% in normalized mean square error (NMSE), 12–15% higher precoding gain, and 25% faster recovery during LoS-to-NLoS transitions compared to baseline clustering methods. Additional evaluations with simulated mmWave channels confirm up to 20% robustness improvement under mobility, highlighting the scalability of the approach. This research establishes adaptive power-aware clustering as a promising enabler for robust, intelligent channel modeling and estimation in next-generation wireless networks.</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

Adaptive power-aware differentiable clustering (APDC) for non-stationary wireless channels with deep learning-based estimation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月3日

查看全文
  • R. Thamilchelvan,
  • C. Gomathy

摘要

Accurate channel estimation in dynamic wireless environments is a critical challenge for emerging 5G and 6G systems, where multipath components (MPCs) vary rapidly due to human mobility, obstacles, and frequent LoS-to-NLoS transitions. Traditional clustering techniques such as K-means or Gaussian mixture models provide useful statistical structures but are limited by fixed cluster assumptions, sensitivity to initialization, and poor adaptability under non-stationary conditions. This research proposes a novel deep learning framework that integrates adaptive, power-aware differentiable clustering with neural channel estimation to overcome these limitations. The method introduces a differentiable clustering layer that assigns MPCs to latent prototypes using power-weighted soft responsibilities, ensuring that high-power paths dominate cluster formation while weaker ones are treated proportionally. Prototypes are updated online with exponential moving averages and forgetting factors, enabling continual adaptation to time-varying environments. Cluster summaries, including prototype means, spreads, and power distributions, are then used as conditioning signals for a neural estimator via cross-attention and feature-wise modulation. The entire model is trained end-to-end to minimize channel estimation error while enforcing compact, stable, and interpretable clusters. Experimental validation with sub-6 GHz USRP measurements in indoor scenarios demonstrates improvements of 18–22% in normalized mean square error (NMSE), 12–15% higher precoding gain, and 25% faster recovery during LoS-to-NLoS transitions compared to baseline clustering methods. Additional evaluations with simulated mmWave channels confirm up to 20% robustness improvement under mobility, highlighting the scalability of the approach. This research establishes adaptive power-aware clustering as a promising enabler for robust, intelligent channel modeling and estimation in next-generation wireless networks.