Joint channel estimation and feedback with masked token transformers in massive MIMO systems.

Area of Science:

• Wireless Communication
• Deep Learning Applications
• Signal Processing

Background:

• Channel estimation and feedback are critical for massive MIMO system performance.
• Existing deep learning methods often neglect intrinsic correlations in channel state information (CSI).
• This oversight leads to suboptimal performance, especially in challenging environments.

Purpose of the Study:

• To propose a novel encoder-decoder network for joint channel estimation and feedback in massive MIMO systems.
• To leverage frequency-domain correlations within CSI for enhanced accuracy.
• To improve operational efficiency and estimation precision.

Main Methods:

• An encoder-decoder network architecture is employed for channel compression.
• A self-mask-attention coding mechanism captures and reconstructs correlation features.
• An active masking strategy enhances operational efficiency.
• A multilayer perceptron denoising module refines channel estimation in the decoder.

Main Results:

• The proposed method demonstrates superior performance in joint channel estimation and feedback tasks compared to state-of-the-art techniques.
• The approach also achieves competitive performance in individual channel estimation and feedback tasks.
• Experimental results validate the effectiveness of leveraging frequency-domain correlations.

Conclusions:

• The developed encoder-decoder network effectively addresses the limitations of existing methods by exploiting CSI correlations.
• The self-mask-attention and denoising modules contribute to improved accuracy and efficiency.
• This work offers a promising direction for advanced channel estimation and feedback in massive MIMO systems.