Passive Beamforming Design of IRS-Assisted MIMO Systems Based on Deep Learning

Area of Science:

• Wireless communication systems
• Signal processing
• Machine learning applications in telecommunications

Background:

• Optimizing passive beamforming in intelligent reflecting surface (IRS)-assisted MIMO systems is critical for maximizing spectral efficiency.
• Existing schemes struggle with the IRS unit-modulus constraint and fail to effectively utilize channel state information (CSI).
• Current methods often treat all input data equally, lacking focus on crucial information, and suffer from high complexity.

Purpose of the Study:

• To develop an advanced scheme for optimizing passive beamforming in IRS-assisted MIMO systems.
• To enhance the exploitation of channel state information (CSI) for improved system performance.
• To reduce the computational complexity and improve the generalization ability of the network.

Main Methods:

• Implementation of a three-channel data input structure.
• Development of an attention mechanism-assisted unsupervised learning scheme.
• Prioritization of key information within input data to enhance network expression and generalization.

Main Results:

• The proposed scheme demonstrates significant improvements in spectral efficiency compared to existing methods.
• A notable reduction in computational complexity was observed.
• The scheme exhibits rapid convergence during simulations.

Conclusions:

• The attention mechanism-assisted unsupervised learning scheme effectively addresses limitations in existing IRS-assisted MIMO systems.
• This approach offers a superior method for optimizing passive beamforming by better leveraging CSI.
• The proposed method provides a computationally efficient and high-performing solution for enhancing spectral efficiency.