DGD-CNet: Denoising Gated Recurrent Unit with a Dropout-Based CSI Network for IRS-Aided Massive MIMO Systems

Area of Science:

• Wireless communication
• Artificial intelligence in telecommunications
• Signal processing

Background:

• Massive MIMO and Intelligent Reflecting Surfaces (IRS) are crucial for 6G networks, especially in Non-Line-of-Sight (NLoS) conditions.
• Passive IRS deployment faces channel estimation challenges in Frequency Division Duplex (FDD)-based Massive MIMO due to high feedback overhead.
• Existing methods struggle to balance feedback reduction and accuracy in complex wireless environments.

Purpose of the Study:

• To introduce a novel deep learning model, Denoising Gated Recurrent Unit with Dropout-based Channel state information Network (DGD-CNet), for efficient channel estimation.
• To address the feedback overhead challenge in FDD-based IRS-aided Massive MIMO systems.
• To enhance channel estimation accuracy and capture spatio-temporal dynamics in time-varying channels.

Main Methods:

• Development of the DGD-CNet model, integrating Gated Recurrent Unit (GRU) with Dropout (DO) for enhanced learning.
• Application of the DGD-CNet model to FDD-based IRS-aided Massive MIMO systems.
• Performance evaluation through Normalized Mean Square Error (NMSE), correlation coefficient, and system accuracy metrics.

Main Results:

• The DGD-CNet model achieved significant improvements over existing methods, with at least a 26% reduction in NMSE.
• A 2% increase in correlation coefficient and a 4% rise in system accuracy were observed under low-compression ratios (Low-CR) in indoor settings.
• The model demonstrated robust performance across various compression ratios and in outdoor scenarios.

Conclusions:

• The proposed DGD-CNet model effectively reduces feedback overhead and enhances channel estimation accuracy in 6G IRS-aided Massive MIMO systems.
• The integration of GRU and DO enables the model to capture complex channel characteristics.
• DGD-CNet offers a promising solution for efficient and accurate channel estimation in future wireless networks.