Scholarlink

<p>The increasing demands of Terahertz (THz) Ultra-Massive MIMO (UM-MIMO) systems present several challenges, including wideband channel estimation, high path loss, and beam squint effects. These systems promise ultra-high data rates for next-generation wireless communications but face difficulties with the measurement matrix and accurate channel state information acquisition, especially under wideband and sparse channel conditions. Existing methods often struggle to balance accuracy, particularly when handling the dual challenges of near-field and far-field propagation in wideband scenarios. This paper presents the Adaptive Sparse Beam dynamic knowledge-driven Dual Coder Feedback Network (ASB-DFNet) for downlink channel estimation in THz UM-MIMO systems. The method begins with a deep autoencoder de-quantization model (DeepAE-DQ) to reduce quantization noise from low-resolution analog-to-digital converters. It incorporates a hybrid wideband dictionary antenna (HWDA) system, dynamically adjusting for far-field (DFT-based HWDA) and near-field (adaptive-driven HWDA) channels, effectively capturing local and distant propagation effects across a broad frequency range. The DFT-based HWDA enhances channel estimation by improving angular resolution and representing off-grid scatterers, minimizing estimation errors. The adaptive-driven HWDA continuously adjusts to near- and far-field characteristics, achieving high accuracy in diverse propagation scenarios. Further, the dynamic knowledge-driven deep prior aided vector learned approximate message passing (DKDPAV-LAMP) model improves channel estimation by leveraging adaptive learning and prior knowledge, addressing issues related to wideband channels and varied propagation conditions. The approach further contains a dual coder feedback scheme in order to minimize feedback overhead as well as improving the quality of wideband channel estimations. During performance assessments, ASB-DFNet exhibited better performance in critical measures like NMSE, BER and SE compared to the current methods at different SNR, bandwidth and antenna arrangement. The results highlight the appropriateness of the method to be used in real-time communication in 6G systems that provide better reliability and accuracy in challenging THz communication conditions.</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

Enhancing wideband channel estimation in THz UM-MIMO: adaptive sparse beam and dual feedback network under hybrid spatial beam squint effect

  • 期刊论文

  • 发表日期：

    2026年3月13日

查看全文
  • S. Ranjith,
  • G. Balachandran,
  • C. H. Venkata Siva Prasad,
  • T. R. Chenthil

摘要

The increasing demands of Terahertz (THz) Ultra-Massive MIMO (UM-MIMO) systems present several challenges, including wideband channel estimation, high path loss, and beam squint effects. These systems promise ultra-high data rates for next-generation wireless communications but face difficulties with the measurement matrix and accurate channel state information acquisition, especially under wideband and sparse channel conditions. Existing methods often struggle to balance accuracy, particularly when handling the dual challenges of near-field and far-field propagation in wideband scenarios. This paper presents the Adaptive Sparse Beam dynamic knowledge-driven Dual Coder Feedback Network (ASB-DFNet) for downlink channel estimation in THz UM-MIMO systems. The method begins with a deep autoencoder de-quantization model (DeepAE-DQ) to reduce quantization noise from low-resolution analog-to-digital converters. It incorporates a hybrid wideband dictionary antenna (HWDA) system, dynamically adjusting for far-field (DFT-based HWDA) and near-field (adaptive-driven HWDA) channels, effectively capturing local and distant propagation effects across a broad frequency range. The DFT-based HWDA enhances channel estimation by improving angular resolution and representing off-grid scatterers, minimizing estimation errors. The adaptive-driven HWDA continuously adjusts to near- and far-field characteristics, achieving high accuracy in diverse propagation scenarios. Further, the dynamic knowledge-driven deep prior aided vector learned approximate message passing (DKDPAV-LAMP) model improves channel estimation by leveraging adaptive learning and prior knowledge, addressing issues related to wideband channels and varied propagation conditions. The approach further contains a dual coder feedback scheme in order to minimize feedback overhead as well as improving the quality of wideband channel estimations. During performance assessments, ASB-DFNet exhibited better performance in critical measures like NMSE, BER and SE compared to the current methods at different SNR, bandwidth and antenna arrangement. The results highlight the appropriateness of the method to be used in real-time communication in 6G systems that provide better reliability and accuracy in challenging THz communication conditions.