Scholarlink

<p>Reconfigurable intelligent surfaces (RIS) is an emerging technology with the remarkable potential to enhance the performance of 6G communication systems. However, the inherent characteristics of the RIS’s passive nature and the numerous unit cells involved introduce unprecedented complexity in channel estimation, consequently exacerbating the pilot overhead. In response to these challenges, we propose a deep unfolding-based channel estimation framework named ADMM-CE-Net, designed for RIS-assisted millimeter-wave (mmWave) massive MIMO communication systems. Capitalizing on the sparsity characteristics of the cascaded channel in the angular domain, we reformulate the channel estimation paradigm as a compressive sensing problem. This reformulation enables us to leverage the mathematical rigor of the alternating direction method of multipliers (ADMM) algorithm, thereby establishing an iterative optimization framework for sparse signal recovery. We transform this iterative optimization procedure into a deep neural network structure, ADMM-CE-Net, by intelligently incorporating learnable parameters that adapt to the channel environment. Experimental results and performance evaluations reveal that our proposed algorithm achieves superior performance metrics compared to the two benchmark approaches, particularly in low signal-to-noise ratio regimes while maintaining comparable estimation accuracy with reduced pilot overhead. Our work may provide an advancement for future practical RIS implementation of 6G 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

Admm-based learnable channel estimation for RIS-aided mmWave massive MIMO systems

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Ying Wu,
  • Ziyan Liu,
  • Shitong Cheng,
  • Banghai He

摘要

Reconfigurable intelligent surfaces (RIS) is an emerging technology with the remarkable potential to enhance the performance of 6G communication systems. However, the inherent characteristics of the RIS’s passive nature and the numerous unit cells involved introduce unprecedented complexity in channel estimation, consequently exacerbating the pilot overhead. In response to these challenges, we propose a deep unfolding-based channel estimation framework named ADMM-CE-Net, designed for RIS-assisted millimeter-wave (mmWave) massive MIMO communication systems. Capitalizing on the sparsity characteristics of the cascaded channel in the angular domain, we reformulate the channel estimation paradigm as a compressive sensing problem. This reformulation enables us to leverage the mathematical rigor of the alternating direction method of multipliers (ADMM) algorithm, thereby establishing an iterative optimization framework for sparse signal recovery. We transform this iterative optimization procedure into a deep neural network structure, ADMM-CE-Net, by intelligently incorporating learnable parameters that adapt to the channel environment. Experimental results and performance evaluations reveal that our proposed algorithm achieves superior performance metrics compared to the two benchmark approaches, particularly in low signal-to-noise ratio regimes while maintaining comparable estimation accuracy with reduced pilot overhead. Our work may provide an advancement for future practical RIS implementation of 6G networks.